- 最新动态
- 专刊征文
- 推荐文章
- 领域热点
-
- Research Progress on Distributed System Model Checking Technologies
- Failure Reproducing Test Case Generation Method Based on Large Language Model
- Survey on Concept-based Modeling Methods for Interpretable Deep Learning
- Survey on Application of Retrieval-augmented Generation in Software Engineering
- More+
- Current Issue
- Online First
- Archive
- Click Rank
- Most Downloaded
- Review Article
- Special Issue Article
- Monographic Series
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2026,37(8):3052-3088, DOI: 10.13328/j.cnki.jos.007595, CSTR: 32375.14.jos.007595
Abstract:
Modeling embedded systems is an essential component of model-based software development. The architecture analysis and design language (AADL), with its ability to formally express hardware-software structures and interaction relationships, is widely applied in system design. Large language models (LLMs) provide a new pathway for generating architecture models from natural language requirements. However, existing approaches exhibit significant limitations in requirement semantic understanding, boundary identification of AADL components, and construction of connection relationships, which constrain their practicality and the quality of generated models. To address these challenges, this study proposes an intelligent modeling approach for embedded systems, termed SmartGen-AADL. The overall framework is built upon a multi-agent collaboration mechanism, integrating key techniques such as semantic parsing, structural recognition, and prompt-enhanced generation, thus enabling high-quality transformation from natural language requirements into structured AADL models. The method consists of three core stages: (1) a structural agent identifies system architectures from system architecture documents and extracts standardized requirement statements; (2) a sub-problem agent performs item-level analysis and interaction mining to refine requirement granularity and explicitly model component interactions; (3) a component generation agent incorporates structural guidance and retrieval-augmented generation (RAG) of similar components into semantic prompts, guiding the LLM to produce component code that conforms to AADL syntax. To support this process, a knowledge base of “itemized requirements-AADL components” and a semantic alignment dataset of “system architecture documents-AADL architectures” are constructed. Experimental results on 15 embedded system application scenarios demonstrate that, compared with approaches solely relying on prompt engineering, the proposed multi-agent collaborative modeling method achieves significant improvements across four mainstream LLMs. Among them, the performance gains are most pronounced on the DeepSeek-r1 model: the component line error rate is reduced by an average of 34.37%, FBERT semantic similarity is increased by 6.21%, structural matching accuracy improves by more than 20%, and human evaluation scores rise by approximately 0.7 points. Furthermore, results from the ablation study reveal that the sub-problem identification mechanism enhances control over modeling granularity. The system structure tree contributes to component organization and hierarchical topology information. The retrieval-augmented generation mechanism supplies external knowledge support and reduces hallucination. Communication connection recognition ensures interface completeness and closed interaction loops. The synergy of these four mechanisms substantially promotes alignment between natural language requirements and the AADL modeling language, thereby improving model consistency.
Modeling embedded systems is an essential component of model-based software development. The architecture analysis and design language (AADL), with its ability to formally express hardware-software structures and interaction relationships, is widely applied in system design. Large language models (LLMs) provide a new pathway for generating architecture models from natural language requirements. However, existing approaches exhibit significant limitations in requirement semantic understanding, boundary identification of AADL components, and construction of connection relationships, which constrain their practicality and the quality of generated models. To address these challenges, this study proposes an intelligent modeling approach for embedded systems, termed SmartGen-AADL. The overall framework is built upon a multi-agent collaboration mechanism, integrating key techniques such as semantic parsing, structural recognition, and prompt-enhanced generation, thus enabling high-quality transformation from natural language requirements into structured AADL models. The method consists of three core stages: (1) a structural agent identifies system architectures from system architecture documents and extracts standardized requirement statements; (2) a sub-problem agent performs item-level analysis and interaction mining to refine requirement granularity and explicitly model component interactions; (3) a component generation agent incorporates structural guidance and retrieval-augmented generation (RAG) of similar components into semantic prompts, guiding the LLM to produce component code that conforms to AADL syntax. To support this process, a knowledge base of “itemized requirements-AADL components” and a semantic alignment dataset of “system architecture documents-AADL architectures” are constructed. Experimental results on 15 embedded system application scenarios demonstrate that, compared with approaches solely relying on prompt engineering, the proposed multi-agent collaborative modeling method achieves significant improvements across four mainstream LLMs. Among them, the performance gains are most pronounced on the DeepSeek-r1 model: the component line error rate is reduced by an average of 34.37%, FBERT semantic similarity is increased by 6.21%, structural matching accuracy improves by more than 20%, and human evaluation scores rise by approximately 0.7 points. Furthermore, results from the ablation study reveal that the sub-problem identification mechanism enhances control over modeling granularity. The system structure tree contributes to component organization and hierarchical topology information. The retrieval-augmented generation mechanism supplies external knowledge support and reduces hallucination. Communication connection recognition ensures interface completeness and closed interaction loops. The synergy of these four mechanisms substantially promotes alignment between natural language requirements and the AADL modeling language, thereby improving model consistency.
2026,37(8):3089-3115, DOI: 10.13328/j.cnki.jos.007593, CSTR: 32375.14.jos.007593
Abstract:
Code generation agents based on large language models (LLMs) are profoundly revolutionizing the software development paradigm. Compared with previous code generation techniques, code generation agents have the following three core features. The first feature is autonomy. The agents can independently execute the entire workflow from task decomposition to coding and debugging. The second is expanded task scope. The agents’ capabilities have extended from generating code snippets to encompassing the full software development life cycle (SDLC). The third is the enhancement of engineering practicality. The research focus has shifted from model algorithmic innovation toward engineering challenges such as process management, system reliability, and tool integration. In recent years, this technical domain has witnessed rapid development and demonstrated tremendous application potential, with explosive growth in related research. To this end, this study presents a systematic review of the field of LLM-based code generation agents. The technology’s developmental trajectory since its inception is traced, and its core techniques including both single-agent and multi-agent systems are sorted out and categorized. Furthermore, this study summarizes both various applications of code generation agents in the full SDLC and the mainstream evaluation benchmarks and metrics, and reviews representative tools. Finally, by analyzing the key challenges, the long-term core research directions in the future for this field are pointed out.
Code generation agents based on large language models (LLMs) are profoundly revolutionizing the software development paradigm. Compared with previous code generation techniques, code generation agents have the following three core features. The first feature is autonomy. The agents can independently execute the entire workflow from task decomposition to coding and debugging. The second is expanded task scope. The agents’ capabilities have extended from generating code snippets to encompassing the full software development life cycle (SDLC). The third is the enhancement of engineering practicality. The research focus has shifted from model algorithmic innovation toward engineering challenges such as process management, system reliability, and tool integration. In recent years, this technical domain has witnessed rapid development and demonstrated tremendous application potential, with explosive growth in related research. To this end, this study presents a systematic review of the field of LLM-based code generation agents. The technology’s developmental trajectory since its inception is traced, and its core techniques including both single-agent and multi-agent systems are sorted out and categorized. Furthermore, this study summarizes both various applications of code generation agents in the full SDLC and the mainstream evaluation benchmarks and metrics, and reviews representative tools. Finally, by analyzing the key challenges, the long-term core research directions in the future for this field are pointed out.
2026,37(8):3116-3143, DOI: 10.13328/j.cnki.jos.007597, CSTR: 32375.14.jos.007597
Abstract:
Large language models (LLMs) have shown great potential in software development question-answering (QA) tasks, providing new approaches for acquiring and understanding code knowledge. However, in complex system software represented by the Linux kernel, the actual capabilities of LLMs in code implementation, understanding key mechanisms, tracing evolutionary history, and analyzing design decisions remain insufficiently validated. Existing benchmarks mainly target general-purpose tasks and suffer from insufficient domain depth, difficulty saturation, and misalignment with real engineering practices, making it difficult to ensure the objectivity, accuracy, and comprehensiveness of domain-specific development knowledge QA. To objectively evaluate the QA capabilities of LLMs in complex system software, this study proposes a benchmark dataset construction method for LLM QA capability evaluation, constructs the high-quality QA benchmark for the Linux kernel (LKQABench), and further designs a multi-judge collaborative code knowledge QA evaluation method (MJ-CCE). LKQABench is built from real technical QA data in developer communities, refined through semantic analysis and human review, resulting in 202 standard QA pairs covering major Linux kernel subsystems and multiple cognitive dimensions. MJ-CCE defines a collaborative scoring and voting mechanism among multiple judge models, evaluating answers across three dimensions: key points coverage, factual correctness, and clarity of expression. Experiments on LKQABench show that current LLMs achieve satisfactory performance on single-point knowledge questions related to kernel implementation but exhibit significant shortcomings, such as missing key points and incomplete reasoning chains, when tackling cross-topic integration, deep reasoning, and version-evolution-related questions. This study not only delineates the capability boundaries of LLMs in software development knowledge QA but also provides empirical evidence to support their continuous optimization in this domain.
Large language models (LLMs) have shown great potential in software development question-answering (QA) tasks, providing new approaches for acquiring and understanding code knowledge. However, in complex system software represented by the Linux kernel, the actual capabilities of LLMs in code implementation, understanding key mechanisms, tracing evolutionary history, and analyzing design decisions remain insufficiently validated. Existing benchmarks mainly target general-purpose tasks and suffer from insufficient domain depth, difficulty saturation, and misalignment with real engineering practices, making it difficult to ensure the objectivity, accuracy, and comprehensiveness of domain-specific development knowledge QA. To objectively evaluate the QA capabilities of LLMs in complex system software, this study proposes a benchmark dataset construction method for LLM QA capability evaluation, constructs the high-quality QA benchmark for the Linux kernel (LKQABench), and further designs a multi-judge collaborative code knowledge QA evaluation method (MJ-CCE). LKQABench is built from real technical QA data in developer communities, refined through semantic analysis and human review, resulting in 202 standard QA pairs covering major Linux kernel subsystems and multiple cognitive dimensions. MJ-CCE defines a collaborative scoring and voting mechanism among multiple judge models, evaluating answers across three dimensions: key points coverage, factual correctness, and clarity of expression. Experiments on LKQABench show that current LLMs achieve satisfactory performance on single-point knowledge questions related to kernel implementation but exhibit significant shortcomings, such as missing key points and incomplete reasoning chains, when tackling cross-topic integration, deep reasoning, and version-evolution-related questions. This study not only delineates the capability boundaries of LLMs in software development knowledge QA but also provides empirical evidence to support their continuous optimization in this domain.
2026,37(8):3144-3160, DOI: 10.13328/j.cnki.jos.007596, CSTR: 32375.14.jos.007596
Abstract:
Large language models (LLMs) have demonstrated remarkable performance in general tasks. However, their trustworthiness, robustness, and applicability in specialized domains remain insufficiently assessed. Using the compilation of software testing textbooks as a representative application scenario, this study constructs 700 carefully designed test questions covering 100 core testing concepts and methods and systematically assesses five representative LLMs in terms of reading comprehension, question-answering (Q&A), and text generation. The experimental results indicate that LLMs generally exhibit strong performance on most questions, achieving high levels of accuracy, completeness, and fluency. However, issues of reliability, such as hallucination and reasoning bias, persist, particularly when addressing current research trends and complex concepts. Further analysis reveals that LLM-generated content provides broader knowledge coverage and greater educational value compared with traditional textbooks, offering effective support for revising and teaching software testing materials. This study not only delineates the specific capability boundaries and typical deficiencies of LLMs in processing domain knowledge but also provides empirical evidence and methodological insights for advancing Q&A-driven intelligent evaluation in professional education and applications.
Large language models (LLMs) have demonstrated remarkable performance in general tasks. However, their trustworthiness, robustness, and applicability in specialized domains remain insufficiently assessed. Using the compilation of software testing textbooks as a representative application scenario, this study constructs 700 carefully designed test questions covering 100 core testing concepts and methods and systematically assesses five representative LLMs in terms of reading comprehension, question-answering (Q&A), and text generation. The experimental results indicate that LLMs generally exhibit strong performance on most questions, achieving high levels of accuracy, completeness, and fluency. However, issues of reliability, such as hallucination and reasoning bias, persist, particularly when addressing current research trends and complex concepts. Further analysis reveals that LLM-generated content provides broader knowledge coverage and greater educational value compared with traditional textbooks, offering effective support for revising and teaching software testing materials. This study not only delineates the specific capability boundaries and typical deficiencies of LLMs in processing domain knowledge but also provides empirical evidence and methodological insights for advancing Q&A-driven intelligent evaluation in professional education and applications.
2026,37(8):3161-3179, DOI: 10.13328/j.cnki.jos.007592, CSTR: 32375.14.jos.007592
Abstract:
Software fault localization is a critical issue in software engineering. In recent years, fault localization methods based on large language models (LLMs) have demonstrated a promising prospect in fault localization tasks. However, existing methods maintain only a single decision path for LLMs, which limits the search scope and results in suboptimal fault localization performance. To this end, this study proposes PRIME, an enhanced fault localization method for LLMs based on parallel exploration. The search scope of LLMs is broadened by designing a parallel exploration mechanism for fault locations. Furthermore, multiple candidate fault locations predicted by LLMs are ranked by combining a node importance evaluation method to generate optimized fault localization results. By conducting comparative analysis with other fault localization methods, comprehensive ablation experiments and parameter influence analysis, it is verified that the proposed method can effectively enhance the fault localization performance of LLMs. Compared with the existing methods, PRIME improves the Top-1 metric by over 18%, and its performance improvements in MAP and MRR metrics can reach 15% and 25%, respectively.
Software fault localization is a critical issue in software engineering. In recent years, fault localization methods based on large language models (LLMs) have demonstrated a promising prospect in fault localization tasks. However, existing methods maintain only a single decision path for LLMs, which limits the search scope and results in suboptimal fault localization performance. To this end, this study proposes PRIME, an enhanced fault localization method for LLMs based on parallel exploration. The search scope of LLMs is broadened by designing a parallel exploration mechanism for fault locations. Furthermore, multiple candidate fault locations predicted by LLMs are ranked by combining a node importance evaluation method to generate optimized fault localization results. By conducting comparative analysis with other fault localization methods, comprehensive ablation experiments and parameter influence analysis, it is verified that the proposed method can effectively enhance the fault localization performance of LLMs. Compared with the existing methods, PRIME improves the Top-1 metric by over 18%, and its performance improvements in MAP and MRR metrics can reach 15% and 25%, respectively.
2026,37(8):3180-3204, DOI: 10.13328/j.cnki.jos.007594, CSTR: 32375.14.jos.007594
Abstract:
Root cause analysis plays a critical role in ensuring the stability and efficiency of modern software systems, particularly in cloud computing and microservice-based systems. Large language models (LLMs), with their powerful natural language processing and data analysis capabilities, have provided new solutions for root cause analysis. LLM-based agents have further enhanced root cause analysis capabilities, such as higher levels of automation and more precise problem localization. While existing research has explored the application of LLMs in root cause analysis, research on LLM-based agents is still at an early stage. To address this gap, this survey provides a comprehensive analysis and summary of current research on LLM-based agents for root cause analysis in cloud computing and microservices systems. The main contents include (1) an overview of the architecture of LLM-based agents and the types of data involved in root cause analysis; (2) a systematic analysis of how LLM-based agents are applied to root cause analysis through the main stages of information collection, root cause localization, and effectiveness evaluation; (3) an exploration of the main challenges and future directions of LLM-based agent technologies in root cause analysis tasks.
Root cause analysis plays a critical role in ensuring the stability and efficiency of modern software systems, particularly in cloud computing and microservice-based systems. Large language models (LLMs), with their powerful natural language processing and data analysis capabilities, have provided new solutions for root cause analysis. LLM-based agents have further enhanced root cause analysis capabilities, such as higher levels of automation and more precise problem localization. While existing research has explored the application of LLMs in root cause analysis, research on LLM-based agents is still at an early stage. To address this gap, this survey provides a comprehensive analysis and summary of current research on LLM-based agents for root cause analysis in cloud computing and microservices systems. The main contents include (1) an overview of the architecture of LLM-based agents and the types of data involved in root cause analysis; (2) a systematic analysis of how LLM-based agents are applied to root cause analysis through the main stages of information collection, root cause localization, and effectiveness evaluation; (3) an exploration of the main challenges and future directions of LLM-based agent technologies in root cause analysis tasks.
2026,37(8):3205-3222, DOI: 10.13328/j.cnki.jos.007598, CSTR: 32375.14.jos.007598
Abstract:
Extracting protocol interactions from textual specification documents written in natural language is useful, especially when to verify the correctness of a protocol before its implementation and application, or when to generate test cases for protocol-connected systems directly from specification documents. Existing approaches for this purpose rely on deep learning or large language models (LLMs). The deep learning approaches require large-scale and high-quality annotated datasets. They may not work well across protocols in different domains due to limitations imposed by the training datasets, and suffer from difficulties in transfer. The LLM-based approaches offer better generalizability, but existing work only uses simple prompt templates. It does not carefully utilize extraction examples in LLM prompting, and the information extraction process lacks optimization, which affects the effectiveness of the proposed approaches. To address these challenges, this study proposes an enhanced LLM-based method for extracting protocol interactions from protocol texts, based on linguistic expression pattern analysis. Specifically, real-world protocol description texts are first analyzed to summarize common linguistic expression patterns in such texts. Then, representative protocol description examples exhibiting these patterns are selected, and corresponding extraction rules are distilled. Further, these examples and rules are integrated to design a rule retrospection chain-of-thought method for LLM-based protocol interaction extraction. Finally, multi-path inference and self-verification techniques are used to optimize the task execution process. Experimental results on multiple protocol datasets show that the proposed method outperforms the baseline methods in terms of precision and recall of protocol interaction extraction, which confirms the effectiveness of the proposed method.
Extracting protocol interactions from textual specification documents written in natural language is useful, especially when to verify the correctness of a protocol before its implementation and application, or when to generate test cases for protocol-connected systems directly from specification documents. Existing approaches for this purpose rely on deep learning or large language models (LLMs). The deep learning approaches require large-scale and high-quality annotated datasets. They may not work well across protocols in different domains due to limitations imposed by the training datasets, and suffer from difficulties in transfer. The LLM-based approaches offer better generalizability, but existing work only uses simple prompt templates. It does not carefully utilize extraction examples in LLM prompting, and the information extraction process lacks optimization, which affects the effectiveness of the proposed approaches. To address these challenges, this study proposes an enhanced LLM-based method for extracting protocol interactions from protocol texts, based on linguistic expression pattern analysis. Specifically, real-world protocol description texts are first analyzed to summarize common linguistic expression patterns in such texts. Then, representative protocol description examples exhibiting these patterns are selected, and corresponding extraction rules are distilled. Further, these examples and rules are integrated to design a rule retrospection chain-of-thought method for LLM-based protocol interaction extraction. Finally, multi-path inference and self-verification techniques are used to optimize the task execution process. Experimental results on multiple protocol datasets show that the proposed method outperforms the baseline methods in terms of precision and recall of protocol interaction extraction, which confirms the effectiveness of the proposed method.
Article Search
Search by issue
Select AllDeselectExport
Display Method:
LIN Hong-Yu, LIU Yang, LUO Hao-Ran, LI Lin-Hai, CAO Hang-Hang, WANG Meng-Na, XU Jia-Hao, LI Quan-Yi, ZHANG Hong-Bin, XING Ming-Jie, WU Yan-Jun
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007649
, CSTR: 32375.14.jos.007649
Abstract:
With the rapid development of deep learning, research on deep learning frameworks and hardware has become a crucial direction for advancing the field. Frameworks provide developers with convenient tools for building deep learning models, while hardware delivers powerful computational capabilities. However, limited adaptability and insufficient compatibility between diverse deep learning frameworks and hardware platforms often lead to performance and scalability issues. To address this challenge, deep learning compilation technology has emerged. Models from different frameworks are efficiently mapped to executable code for specific backend devices through a series of intermediate representations and automated transformations. During this process, various deep learning compilation optimization techniques, such as operator fusion, memory optimization, and auto-tuning, are applied. These not only resolve scalability issues but also significantly improve the computational efficiency of the models. This study first reviews the basic concepts and overall process of deep learning compilation, then summarizes and categorizes the common optimization techniques in deep learning compilation, and finally discusses the challenges faced by the field and potential future development directions.
With the rapid development of deep learning, research on deep learning frameworks and hardware has become a crucial direction for advancing the field. Frameworks provide developers with convenient tools for building deep learning models, while hardware delivers powerful computational capabilities. However, limited adaptability and insufficient compatibility between diverse deep learning frameworks and hardware platforms often lead to performance and scalability issues. To address this challenge, deep learning compilation technology has emerged. Models from different frameworks are efficiently mapped to executable code for specific backend devices through a series of intermediate representations and automated transformations. During this process, various deep learning compilation optimization techniques, such as operator fusion, memory optimization, and auto-tuning, are applied. These not only resolve scalability issues but also significantly improve the computational efficiency of the models. This study first reviews the basic concepts and overall process of deep learning compilation, then summarizes and categorizes the common optimization techniques in deep learning compilation, and finally discusses the challenges faced by the field and potential future development directions.
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007651
, CSTR: 32375.14.jos.007651
Abstract:
Intrusion detection systems rely on prior knowledge, and their input data contain rich semantic information, making them well suited for large language models (LLMs). However, the computational overhead of LLMs and data privacy constraints make their direct deployment difficult. Meanwhile, although small models are convenient for deployment, their performance is limited. To address this issue, this study proposes an iterative model for intrusion detection based on data-driven distillation and performance motivation, termed as RADOM. RADOM performs knowledge distillation through data exchange between LLMs and small models, and improves the quality of LLM-generated data by using incorrect predictions from the small models as iterative feedback. At the same time, this study introduces a feature dimension optimization mechanism, in which LLMs are used to perform feature selection and iterative updating, thereby further improving the classification capability of small models. Experimental results on public datasets and self-collected datasets show that RADOM effectively improves the detection and classification performance of small models for attack behaviors, and the accuracy of small models increases from 67.30% to 96.19%, verifying the effectiveness of the proposed method.
Intrusion detection systems rely on prior knowledge, and their input data contain rich semantic information, making them well suited for large language models (LLMs). However, the computational overhead of LLMs and data privacy constraints make their direct deployment difficult. Meanwhile, although small models are convenient for deployment, their performance is limited. To address this issue, this study proposes an iterative model for intrusion detection based on data-driven distillation and performance motivation, termed as RADOM. RADOM performs knowledge distillation through data exchange between LLMs and small models, and improves the quality of LLM-generated data by using incorrect predictions from the small models as iterative feedback. At the same time, this study introduces a feature dimension optimization mechanism, in which LLMs are used to perform feature selection and iterative updating, thereby further improving the classification capability of small models. Experimental results on public datasets and self-collected datasets show that RADOM effectively improves the detection and classification performance of small models for attack behaviors, and the accuracy of small models increases from 67.30% to 96.19%, verifying the effectiveness of the proposed method.
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007695
, CSTR: 32375.14.jos.007695
Abstract:
Cardinality estimation is a core component of the query optimizer in database management systems (DBMS), and its accuracy directly affects the quality of execution plans. Although existing learning-based cardinality estimation methods outperform traditional methods in certain scenarios, they struggle to maintain consistently high accuracy on highly heterogeneous query workloads, where queries may differ significantly in the number of tables, join patterns, and predicates. Moreover, existing multi-model approaches lack a fine-grained mechanism for selecting models at the query level. Therefore, this study proposes a query routing-based cardinality estimation method, QRCE, whose core ideas include 1) constructing a heterogeneous and scalable Seq2Seq model space, integrating multiple sequence modeling architectures with different structural inductive biases to form a candidate model set; 2) building a gated mixture query router, GMQR, to achieve fine-grained model routing based on query semantics and enable adaptive decision-making for individual queries; 3) proposing a Q-error-based (1+ε)-approximate label strategy, transforming model selection into a supervised learning task, enabling the system to adaptively select the most suitable model based on the semantic features of each query. Experimental results on three benchmark datasets, STATS, JOB-light, and TPC-H, show that QRCE achieves significantly better overall and tail error performance than most baseline methods. Particularly on the structurally complex STATS dataset, QRCE reduces the Q99 error by approximately 53%, effectively avoiding long-tail catastrophic bias. On JOB-light, its Q99 is close to that of the best single model, while maintaining advantages at Q50/Q90/Q95. On TPC-H, the Q99 error is further reduced by about 16%. QRCE demonstrates stronger robustness and adaptability in long-tail queries and complex join scenarios.
Cardinality estimation is a core component of the query optimizer in database management systems (DBMS), and its accuracy directly affects the quality of execution plans. Although existing learning-based cardinality estimation methods outperform traditional methods in certain scenarios, they struggle to maintain consistently high accuracy on highly heterogeneous query workloads, where queries may differ significantly in the number of tables, join patterns, and predicates. Moreover, existing multi-model approaches lack a fine-grained mechanism for selecting models at the query level. Therefore, this study proposes a query routing-based cardinality estimation method, QRCE, whose core ideas include 1) constructing a heterogeneous and scalable Seq2Seq model space, integrating multiple sequence modeling architectures with different structural inductive biases to form a candidate model set; 2) building a gated mixture query router, GMQR, to achieve fine-grained model routing based on query semantics and enable adaptive decision-making for individual queries; 3) proposing a Q-error-based (1+ε)-approximate label strategy, transforming model selection into a supervised learning task, enabling the system to adaptively select the most suitable model based on the semantic features of each query. Experimental results on three benchmark datasets, STATS, JOB-light, and TPC-H, show that QRCE achieves significantly better overall and tail error performance than most baseline methods. Particularly on the structurally complex STATS dataset, QRCE reduces the Q99 error by approximately 53%, effectively avoiding long-tail catastrophic bias. On JOB-light, its Q99 is close to that of the best single model, while maintaining advantages at Q50/Q90/Q95. On TPC-H, the Q99 error is further reduced by about 16%. QRCE demonstrates stronger robustness and adaptability in long-tail queries and complex join scenarios.
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007696
, CSTR: 32375.14.jos.007696
Abstract:
As deep learning models are increasingly deployed in resource-constrained scenarios, model compression techniques such as pruning, quantization, and knowledge distillation have become key approaches for improving model efficiency and deployability. However, while model compression offers advantages in performance and efficiency, it is also intricately linked to security and privacy issues. On the one hand, compression may alter a model’s representation space or decision boundary, thus introducing new security risks or amplifying existing threats. On the other hand, security and privacy requirements have also shaped the development of model compression techniques, making security-aware compression an increasingly important research direction. This study systematically reviews recent progress in model compression research from a security perspective, focusing on two main aspects. First, it examines the security and privacy risks faced by compressed models, including backdoor attacks, adversarial examples, privacy leakage, and emerging risks in frontier application scenarios. Second, it surveys security-aware model compression methods for adversarial robustness and privacy protection, summarizing representative studies that incorporate security constraints and privacy-preserving mechanisms into the compression process. In addition, this study identifies key issues in security and privacy research on compressed models, provides risk-aware guidance for typical deployment scenarios, recommends benchmark-based evaluation dimensions for assessing the security of compressed models, and discusses future research directions in terms of underlying mechanism analysis and security protection in real-world deployments.
As deep learning models are increasingly deployed in resource-constrained scenarios, model compression techniques such as pruning, quantization, and knowledge distillation have become key approaches for improving model efficiency and deployability. However, while model compression offers advantages in performance and efficiency, it is also intricately linked to security and privacy issues. On the one hand, compression may alter a model’s representation space or decision boundary, thus introducing new security risks or amplifying existing threats. On the other hand, security and privacy requirements have also shaped the development of model compression techniques, making security-aware compression an increasingly important research direction. This study systematically reviews recent progress in model compression research from a security perspective, focusing on two main aspects. First, it examines the security and privacy risks faced by compressed models, including backdoor attacks, adversarial examples, privacy leakage, and emerging risks in frontier application scenarios. Second, it surveys security-aware model compression methods for adversarial robustness and privacy protection, summarizing representative studies that incorporate security constraints and privacy-preserving mechanisms into the compression process. In addition, this study identifies key issues in security and privacy research on compressed models, provides risk-aware guidance for typical deployment scenarios, recommends benchmark-based evaluation dimensions for assessing the security of compressed models, and discusses future research directions in terms of underlying mechanism analysis and security protection in real-world deployments.
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007704
, CSTR: 32375.14.jos.007704
Abstract:
Large language models (LLMs) perform well in complex reasoning tasks involving logical operations. However, the internal mechanisms underlying their logical reasoning capabilities remain insufficiently interpretable. Currently, it remains unclear whether neuron subsets closely associated with logical reasoning performance exist within LLMs, what forms of computational operations are performed by these neurons in the network, and how they influence model performance across different tasks. To investigate these key questions concerning the nature of LLM reasoning mechanisms, this study proposes an “identification-intervention-evaluation” framework aimed at systematically analyzing the neuronal operational basis and functional implementation of logical reasoning. A neuron localization method based on gradient importance is used to identify “logic neurons” that are highly sensitive to logical reasoning tasks, and targeted interventions are performed on their activation values to further evaluate their impact on model behavior. The results show that the identified logic-related neuron subsets are mainly concentrated in the middle layers of the network and predominantly located in feed-forward network (FFN) modules. Further analysis indicates that different logical tasks involve both shared neuron components and task-specific differences. Based on these findings, this study explores a lightweight targeted enhancement method for logical reasoning capabilities without modifying model weights. Experimental results show that this method improves the average accuracy of the model on long-chain reasoning tasks in LogicBench by 9.2% and improves performance on logic-sensitive general tasks by 4.57%.
Large language models (LLMs) perform well in complex reasoning tasks involving logical operations. However, the internal mechanisms underlying their logical reasoning capabilities remain insufficiently interpretable. Currently, it remains unclear whether neuron subsets closely associated with logical reasoning performance exist within LLMs, what forms of computational operations are performed by these neurons in the network, and how they influence model performance across different tasks. To investigate these key questions concerning the nature of LLM reasoning mechanisms, this study proposes an “identification-intervention-evaluation” framework aimed at systematically analyzing the neuronal operational basis and functional implementation of logical reasoning. A neuron localization method based on gradient importance is used to identify “logic neurons” that are highly sensitive to logical reasoning tasks, and targeted interventions are performed on their activation values to further evaluate their impact on model behavior. The results show that the identified logic-related neuron subsets are mainly concentrated in the middle layers of the network and predominantly located in feed-forward network (FFN) modules. Further analysis indicates that different logical tasks involve both shared neuron components and task-specific differences. Based on these findings, this study explores a lightweight targeted enhancement method for logical reasoning capabilities without modifying model weights. Experimental results show that this method improves the average accuracy of the model on long-chain reasoning tasks in LogicBench by 9.2% and improves performance on logic-sensitive general tasks by 4.57%.
Available online: August 12, 2026
, DOI: 10.13328/j.cnki.jos.007706
, CSTR: 32375.14.jos.007706
Abstract:
The dynamic single-source shortest paths problem is studied in the massively parallel computation (MPC) model, where the memory size of each machine is $ \mathrm{O}({n}^{\alpha }) $. Here, $ n $ is the number of vertices in the graph, and $ \alpha \in (0, 1) $ is a constant. This study primarily considers directed or undirected graphs with integer edge weights, where the ratio between the maximum and minimum edge weights is upper-bounded by $ \left\lceil poly(\log n)\right\rceil $. The graph update operations include single-edge insertion, deletion, and weight change. To address this problem, this study combines the polynomial matrix inverse method, the path decomposition strategy, and parallel linear algebra algorithms. In addition, by designing an efficient parallel polynomial multiplication algorithm, a randomized parallel dynamic single-source shortest paths algorithm is proposed. The algorithm achieves an update round complexity of $ \mathrm{O}({\alpha }^{-1}\log n) $ and a total memory requirement of $ \tilde{\mathrm{O}}({n}^{3-3(3-\omega )/(4-\alpha /2)-\alpha (\omega /2-1)}) $. Here, the update round complexity refers to the number of synchronous iteration rounds required by the algorithm to recompute, after each edge change in the graph, the distances from the source to other affected vertices. $ \tilde{\mathrm{O}}(\cdot ) $ suppresses polylogarithmic factors, and $ \omega $ represents the exponent of the arithmetic complexity of fast matrix multiplication on matrices of size $ \mathrm{O}({n}^{\alpha }) $. Compared with existing parallel static single-source shortest paths algorithms that require at least $ poly(\log n) $ round complexity, the proposed algorithm significantly reduces the round complexity.
The dynamic single-source shortest paths problem is studied in the massively parallel computation (MPC) model, where the memory size of each machine is $ \mathrm{O}({n}^{\alpha }) $. Here, $ n $ is the number of vertices in the graph, and $ \alpha \in (0, 1) $ is a constant. This study primarily considers directed or undirected graphs with integer edge weights, where the ratio between the maximum and minimum edge weights is upper-bounded by $ \left\lceil poly(\log n)\right\rceil $. The graph update operations include single-edge insertion, deletion, and weight change. To address this problem, this study combines the polynomial matrix inverse method, the path decomposition strategy, and parallel linear algebra algorithms. In addition, by designing an efficient parallel polynomial multiplication algorithm, a randomized parallel dynamic single-source shortest paths algorithm is proposed. The algorithm achieves an update round complexity of $ \mathrm{O}({\alpha }^{-1}\log n) $ and a total memory requirement of $ \tilde{\mathrm{O}}({n}^{3-3(3-\omega )/(4-\alpha /2)-\alpha (\omega /2-1)}) $. Here, the update round complexity refers to the number of synchronous iteration rounds required by the algorithm to recompute, after each edge change in the graph, the distances from the source to other affected vertices. $ \tilde{\mathrm{O}}(\cdot ) $ suppresses polylogarithmic factors, and $ \omega $ represents the exponent of the arithmetic complexity of fast matrix multiplication on matrices of size $ \mathrm{O}({n}^{\alpha }) $. Compared with existing parallel static single-source shortest paths algorithms that require at least $ poly(\log n) $ round complexity, the proposed algorithm significantly reduces the round complexity.
Available online: July 22, 2026
, DOI: 10.13328/j.cnki.jos.007708
, CSTR: 32375.14.jos.007708
Abstract:
With the rapid development of diffusion models and generative adversarial networks, AI-generated images are increasingly used in content creation, while also posing new challenges to content trustworthiness and source attribution. Compared with merely distinguishing authentic images from fake ones, accurately identifying the source generative model of generated images has greater practical significance. For instance, this capability can provide important technical support for content source identification and trustworthy governance in scenarios such as generated content platform regulation, tracing of maliciously forged content, and responsibility tracking of generative models. During the synthesis process, different generative models often leave fine-grained high-frequency artifacts at the pixel level, while also producing model-specific differences in generative pattern consistency at the structural and global levels. This cross-scale artifact phenomenon indicates that existing methods relying solely on single-scale or single-semantic-level feature representations are insufficient for simultaneously capturing fine-grained artifacts and global generative style differences, thus limiting attribution performance in complex multi-model scenarios. To address these issues, this study proposes a collaborative pixel-block artifact modeling framework (CPB-AM) for AI-generated image model source attribution. The proposed method adopts a dual-branch architecture to jointly model generative artifacts at different levels. The pixel-level branch extracts local forgery traces through high-frequency enhancement operations and learns fine-grained artifact features associated with the generative domain. The block-level branch partitions the image into local blocks and models inter-block correlations through a frequency-aware attention aggregation mechanism. This attention mechanism adaptively aggregates low-frequency smooth components and explicitly separates structural consistency representations from residual components, enabling more effective characterization of differences among generative models in terms of spatial structure and overall style. Furthermore, for cross-scale fusion, a cross-branch attention-guided fusion module is designed as a guided feature fusion mechanism. By constructing modulation weights from pixel-level artifact features for block-level representations, the module enables fine-grained residual information to guide and constrain the structural modeling process, thus enhancing the complementarity of cross-scale artifact information. Experimental results on datasets containing real images and multiple generative models demonstrate that the proposed method achieves superior classification performance compared with existing approaches in closed-set generative model source attribution tasks. Under open-set experimental settings involving unseen generative models, the proposed method also exhibits strong robustness in both unknown model rejection and attribution discrimination. These results verify the effectiveness of the collaborative pixel-block artifact modeling strategy in improving generative model source attribution accuracy and adaptability to open scenarios.
With the rapid development of diffusion models and generative adversarial networks, AI-generated images are increasingly used in content creation, while also posing new challenges to content trustworthiness and source attribution. Compared with merely distinguishing authentic images from fake ones, accurately identifying the source generative model of generated images has greater practical significance. For instance, this capability can provide important technical support for content source identification and trustworthy governance in scenarios such as generated content platform regulation, tracing of maliciously forged content, and responsibility tracking of generative models. During the synthesis process, different generative models often leave fine-grained high-frequency artifacts at the pixel level, while also producing model-specific differences in generative pattern consistency at the structural and global levels. This cross-scale artifact phenomenon indicates that existing methods relying solely on single-scale or single-semantic-level feature representations are insufficient for simultaneously capturing fine-grained artifacts and global generative style differences, thus limiting attribution performance in complex multi-model scenarios. To address these issues, this study proposes a collaborative pixel-block artifact modeling framework (CPB-AM) for AI-generated image model source attribution. The proposed method adopts a dual-branch architecture to jointly model generative artifacts at different levels. The pixel-level branch extracts local forgery traces through high-frequency enhancement operations and learns fine-grained artifact features associated with the generative domain. The block-level branch partitions the image into local blocks and models inter-block correlations through a frequency-aware attention aggregation mechanism. This attention mechanism adaptively aggregates low-frequency smooth components and explicitly separates structural consistency representations from residual components, enabling more effective characterization of differences among generative models in terms of spatial structure and overall style. Furthermore, for cross-scale fusion, a cross-branch attention-guided fusion module is designed as a guided feature fusion mechanism. By constructing modulation weights from pixel-level artifact features for block-level representations, the module enables fine-grained residual information to guide and constrain the structural modeling process, thus enhancing the complementarity of cross-scale artifact information. Experimental results on datasets containing real images and multiple generative models demonstrate that the proposed method achieves superior classification performance compared with existing approaches in closed-set generative model source attribution tasks. Under open-set experimental settings involving unseen generative models, the proposed method also exhibits strong robustness in both unknown model rejection and attribution discrimination. These results verify the effectiveness of the collaborative pixel-block artifact modeling strategy in improving generative model source attribution accuracy and adaptability to open scenarios.
Available online: July 22, 2026
, DOI: 10.13328/j.cnki.jos.007703
, CSTR: 32375.14.jos.007703
Abstract:
Geological disaster knowledge services play a critical role in emergency command, public safety, and science communication. However, existing large language model (LLM)-based question-answering systems still have limitations in multi-source heterogeneous knowledge fusion, complex task planning, and highly trustworthy reasoning. To address these issues, this study proposes a multi-agent collaborative reasoning framework for geological disaster knowledge services. Driven by explicit task states, the proposed framework decomposes complex queries into subtasks such as retrieval, analysis, generation, and verification through a role-based agent collaboration mechanism and ensures the traceability of the reasoning process and the controllability of results through task write-back and closed-loop consistency verification. At the data level, the framework supports unified semantic access to knowledge from multiple sources, such as popular science and domain-specific knowledge, laws and regulations, and disaster records. At the reasoning level, a hierarchical reasoning process based on collaborative execution and quality control is constructed. Experimental results demonstrate that the proposed framework outperforms the baseline method, which is based on a single model combined with external retrieval, in terms of domain knowledge coverage, answer consistency, and collaborative reasoning capability, and shows more stable performance in high-reliability tasks involving laws, regulations, and disaster statistics. This study provides a feasible systematic solution for the application of multi-agent LLMs in disaster emergency knowledge services.
Geological disaster knowledge services play a critical role in emergency command, public safety, and science communication. However, existing large language model (LLM)-based question-answering systems still have limitations in multi-source heterogeneous knowledge fusion, complex task planning, and highly trustworthy reasoning. To address these issues, this study proposes a multi-agent collaborative reasoning framework for geological disaster knowledge services. Driven by explicit task states, the proposed framework decomposes complex queries into subtasks such as retrieval, analysis, generation, and verification through a role-based agent collaboration mechanism and ensures the traceability of the reasoning process and the controllability of results through task write-back and closed-loop consistency verification. At the data level, the framework supports unified semantic access to knowledge from multiple sources, such as popular science and domain-specific knowledge, laws and regulations, and disaster records. At the reasoning level, a hierarchical reasoning process based on collaborative execution and quality control is constructed. Experimental results demonstrate that the proposed framework outperforms the baseline method, which is based on a single model combined with external retrieval, in terms of domain knowledge coverage, answer consistency, and collaborative reasoning capability, and shows more stable performance in high-reliability tasks involving laws, regulations, and disaster statistics. This study provides a feasible systematic solution for the application of multi-agent LLMs in disaster emergency knowledge services.
Available online: July 22, 2026
, DOI: 10.13328/j.cnki.jos.007646
, CSTR: 32375.14.jos.007646
Abstract:
Double-block-length hash functions are a classical approach for amplifying the concrete security of hash functions. This construction has been proven to achieve optimal quantum collision resistance under certain conditions. However, whether double-block-length constructions can still achieve optimal concrete security in the stronger and more applicable quantum security model of the collapsing property remains an open question. To conduct a further study on this issue, this study considers the collapsing property, which extends the notion of collision resistance in the quantum setting. This study focuses on the collapsing security of Nandi’s double-block-length construction based on a random oracle. This study proposes that when the permutations $ \pi $ within the construction is composed completely of a number of c-cycle permutations (i.e., for any $ {x}\in{{\{0, 1\}}}^{{m}} $, $ \pi^{{a}}{(x) =x} $ if and only if $ {a=c} $), the collapsing security of this construction is optimal. Constructing a permutation $ \pi $ composed solely of c-cycle permutations is straightforward. Optimal collapsing security implies that when the output size of the random oracle is n bits, the adversary can effectively distinguish between the two states, measuring the hash value of a quantum superposition of messages and measuring the message superposition itself, only after making at least $ \text{O(}{{2}}^{{2n/3}}\text{)} $ queries. The proposed optimal construction can also be extended by the Merkle-Damg?rd construction. The extended hash function retains the collapsing property. Therefore, this study provides a theoretical foundation for the design of collapsing hash functions in the future.
Double-block-length hash functions are a classical approach for amplifying the concrete security of hash functions. This construction has been proven to achieve optimal quantum collision resistance under certain conditions. However, whether double-block-length constructions can still achieve optimal concrete security in the stronger and more applicable quantum security model of the collapsing property remains an open question. To conduct a further study on this issue, this study considers the collapsing property, which extends the notion of collision resistance in the quantum setting. This study focuses on the collapsing security of Nandi’s double-block-length construction based on a random oracle. This study proposes that when the permutations $ \pi $ within the construction is composed completely of a number of c-cycle permutations (i.e., for any $ {x}\in{{\{0, 1\}}}^{{m}} $, $ \pi^{{a}}{(x) =x} $ if and only if $ {a=c} $), the collapsing security of this construction is optimal. Constructing a permutation $ \pi $ composed solely of c-cycle permutations is straightforward. Optimal collapsing security implies that when the output size of the random oracle is n bits, the adversary can effectively distinguish between the two states, measuring the hash value of a quantum superposition of messages and measuring the message superposition itself, only after making at least $ \text{O(}{{2}}^{{2n/3}}\text{)} $ queries. The proposed optimal construction can also be extended by the Merkle-Damg?rd construction. The extended hash function retains the collapsing property. Therefore, this study provides a theoretical foundation for the design of collapsing hash functions in the future.
Available online: July 15, 2026
, DOI: 10.13328/j.cnki.jos.007682
, CSTR: 32375.14.jos.007682
Abstract:
With the increasing complexity of cyber attacks, the collection and integration of threat intelligence face challenges of fragmentation. In this context, enterprises have begun building private threat intelligence knowledge graph systems. However, traditional construction methods suffer from inefficiency and limited effectiveness due to inaccurate extraction of entities and their relationships. To address these issues, this study proposes OPFA: a general threat intelligence knowledge graph construction framework based on model context protocol (MCP) agents. Through dynamic semantic subject positioning, large language model prompt templates are fine-tuned to drive agents to extract threat entities and their relationships. These threat entities and relationships are automatically created as nodes and edges in the knowledge graph, and threat intelligence is then linked through entity attribute value matching to form a complete knowledge graph. The system subsequently retrieves MCP resources for critical threat entities (such as vulnerabilities and malicious samples) to achieve customized knowledge graph expansion. Experimental results demonstrate that the agents not only improve Precision (97.22% and 97.83%, respectively), Recall (90.91% and 95.52%, respectively), and F1-score (94.44% and 96.66%, respectively) in threat entity and entity relationship extraction, but also effectively enhance the construction efficiency of the knowledge graph.
With the increasing complexity of cyber attacks, the collection and integration of threat intelligence face challenges of fragmentation. In this context, enterprises have begun building private threat intelligence knowledge graph systems. However, traditional construction methods suffer from inefficiency and limited effectiveness due to inaccurate extraction of entities and their relationships. To address these issues, this study proposes OPFA: a general threat intelligence knowledge graph construction framework based on model context protocol (MCP) agents. Through dynamic semantic subject positioning, large language model prompt templates are fine-tuned to drive agents to extract threat entities and their relationships. These threat entities and relationships are automatically created as nodes and edges in the knowledge graph, and threat intelligence is then linked through entity attribute value matching to form a complete knowledge graph. The system subsequently retrieves MCP resources for critical threat entities (such as vulnerabilities and malicious samples) to achieve customized knowledge graph expansion. Experimental results demonstrate that the agents not only improve Precision (97.22% and 97.83%, respectively), Recall (90.91% and 95.52%, respectively), and F1-score (94.44% and 96.66%, respectively) in threat entity and entity relationship extraction, but also effectively enhance the construction efficiency of the knowledge graph.
Available online: July 15, 2026
, DOI: 10.13328/j.cnki.jos.007679
, CSTR: 32375.14.jos.007679
Abstract:
Solid-state drives (SSDs), owing to their high read/write performance and low power consumption, have been widely used in diverse scenarios such as embedded systems and data centers. Among them, the garbage collection (GC) mechanism plays a critical role in space reclamation efficiency and device lifespan. However, existing GC methods still suffer from limitations in triggering flexibility, data migration efficiency, and wear leveling effectiveness. To enhance the adaptability and reliability of GC under varying workloads, this study proposes a write pattern-aware and heat-tiered garbage collection (WHGC) method. WHGC first constructs a write pattern-aware indicator and designs a multi-threshold triggering mechanism to enable dynamic GC scheduling under both idle and busy system states. It then introduces a block scoring strategy that integrates space utilization and wear level to guide victim block selection. During the valid page migration phase, WHGC employs an exponential decay model to identify the heat of logical pages, and combines it with block erase count classification to establish a mapping between heat levels and wear levels, thus guiding valid pages to migrate to blocks of corresponding levels. Experimental results based on the SSDsim platform demonstrate that WHGC achieves a longer device lifespan than other typical GC methods across all datasets, while maintaining superior overall performance in terms of write amplification factor, the number of migrated valid pages, and the standard deviation of erase counts.
Solid-state drives (SSDs), owing to their high read/write performance and low power consumption, have been widely used in diverse scenarios such as embedded systems and data centers. Among them, the garbage collection (GC) mechanism plays a critical role in space reclamation efficiency and device lifespan. However, existing GC methods still suffer from limitations in triggering flexibility, data migration efficiency, and wear leveling effectiveness. To enhance the adaptability and reliability of GC under varying workloads, this study proposes a write pattern-aware and heat-tiered garbage collection (WHGC) method. WHGC first constructs a write pattern-aware indicator and designs a multi-threshold triggering mechanism to enable dynamic GC scheduling under both idle and busy system states. It then introduces a block scoring strategy that integrates space utilization and wear level to guide victim block selection. During the valid page migration phase, WHGC employs an exponential decay model to identify the heat of logical pages, and combines it with block erase count classification to establish a mapping between heat levels and wear levels, thus guiding valid pages to migrate to blocks of corresponding levels. Experimental results based on the SSDsim platform demonstrate that WHGC achieves a longer device lifespan than other typical GC methods across all datasets, while maintaining superior overall performance in terms of write amplification factor, the number of migrated valid pages, and the standard deviation of erase counts.
Available online: July 15, 2026
, DOI: 10.13328/j.cnki.jos.007680
, CSTR: 32375.14.jos.007680
Abstract:
With the continuous growth of data scale and query complexity, traditional database systems exhibit significant redundant computation and resource contention when handling highly concurrent and diverse query workloads. Multi-query optimization (MQO) is a crucial approach for enhancing database performance. Its core principle is to identify and share overlapping computations among different queries to minimize repetitive execution and optimize resource utilization. Based on the timing and granularity of reuse, existing research is primarily divided into two technical routes: computation-reuse-based MQO and storage-reuse-based MQO. The former implements operator-level or plan-level sharing during query execution, while the latter achieves persistent reuse at the storage layer, also known as materialized view technology. With the advancement of artificial intelligence, reinforcement learning and deep learning methods are widely used for benefit estimation, plan generation, and sharing strategy decision-making, enabling the computation reuse process to achieve self-awareness, self-optimization, and self-evolution. This study systematically reviews the evolution from traditional methods to AI-enabled approaches, compares and summarizes the optimization objectives, decision mechanisms, and system architectures of various models, and analyzes the limitations of existing methods in terms of generality, scalability, and collaborative optimization. Finally, future research directions for intelligent and multi-level collaborative database optimization are envisioned.
With the continuous growth of data scale and query complexity, traditional database systems exhibit significant redundant computation and resource contention when handling highly concurrent and diverse query workloads. Multi-query optimization (MQO) is a crucial approach for enhancing database performance. Its core principle is to identify and share overlapping computations among different queries to minimize repetitive execution and optimize resource utilization. Based on the timing and granularity of reuse, existing research is primarily divided into two technical routes: computation-reuse-based MQO and storage-reuse-based MQO. The former implements operator-level or plan-level sharing during query execution, while the latter achieves persistent reuse at the storage layer, also known as materialized view technology. With the advancement of artificial intelligence, reinforcement learning and deep learning methods are widely used for benefit estimation, plan generation, and sharing strategy decision-making, enabling the computation reuse process to achieve self-awareness, self-optimization, and self-evolution. This study systematically reviews the evolution from traditional methods to AI-enabled approaches, compares and summarizes the optimization objectives, decision mechanisms, and system architectures of various models, and analyzes the limitations of existing methods in terms of generality, scalability, and collaborative optimization. Finally, future research directions for intelligent and multi-level collaborative database optimization are envisioned.
Available online: July 15, 2026
, DOI: 10.13328/j.cnki.jos.007630
, CSTR: 32375.14.jos.007630
Abstract:
Micro-expressions can reveal genuine emotions that individuals attempt to conceal and therefore have significant application value in areas such as lie detection, healthcare, and fine-grained emotion recognition. Unlike macro-expressions, micro-expressions are brief and subtle facial movements. In addition, the construction of micro-expression datasets involves high costs and considerable effort, which results in limited data availability and makes it difficult for models to capture subtle and transient facial changes. Micro-expression analysis mainly consists of two important research branches: micro-expression recognition and micro-expression detection. Currently, deep learning-based methods dominate research in this field. Accordingly, this study comprehensively reviews representative works and recent advances in deep learning-based micro-expression recognition and micro-expression detection. Firstly, the study provides a systematic summary of commonly used open-source micro-expression datasets. Secondly, it separately introduces representative research on micro-expression recognition (MER) and micro-expression detection (MED), classifies and organizes these methods, and discusses their respective advantages and limitations. Furthermore, it presents commonly used evaluation protocols and metrics for micro-expression analysis and integrates discussions on the performance of various methods. Finally, it summarizes the overall development of micro-expression analysis and offers insights into future research directions.
Micro-expressions can reveal genuine emotions that individuals attempt to conceal and therefore have significant application value in areas such as lie detection, healthcare, and fine-grained emotion recognition. Unlike macro-expressions, micro-expressions are brief and subtle facial movements. In addition, the construction of micro-expression datasets involves high costs and considerable effort, which results in limited data availability and makes it difficult for models to capture subtle and transient facial changes. Micro-expression analysis mainly consists of two important research branches: micro-expression recognition and micro-expression detection. Currently, deep learning-based methods dominate research in this field. Accordingly, this study comprehensively reviews representative works and recent advances in deep learning-based micro-expression recognition and micro-expression detection. Firstly, the study provides a systematic summary of commonly used open-source micro-expression datasets. Secondly, it separately introduces representative research on micro-expression recognition (MER) and micro-expression detection (MED), classifies and organizes these methods, and discusses their respective advantages and limitations. Furthermore, it presents commonly used evaluation protocols and metrics for micro-expression analysis and integrates discussions on the performance of various methods. Finally, it summarizes the overall development of micro-expression analysis and offers insights into future research directions.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007628
, CSTR: 32375.14.jos.007628
Abstract:
Improving robustness to feature noise is one of the most challenging problems in multimodal sentiment analysis (MSA). Recent studies have proposed efficient MSA models that consider missing modalities, but they typically focus on specific types of defects, which leads to limited robustness and generalization in real-world scenarios where multiple types of noise coexist. In addition, deep interactions across different modalities are often inadequately modeled, making it difficult to fully capture multimodal emotional semantics. To address these issues, this study proposes a robust MSA method that integrates adversarial learning with a multi-view network, namely ALMV. Specifically, temporal modality feature masking is adopted to simulate noisy data, and noise-original instance pairs are constructed with intact sequences for data augmentation. Secondly, temporal convolutional networks and Transformer encoders are used to extract local and global information from every modality sequence, and a weight-controlled multi-view network is constructed to learn joint multimodal representations from the noise-original instance pairs. Additionally, a novel multi-level adversarial training strategy with semantic reconstruction supervision is introduced to learn unified representations between noisy and complete data at both the modality level and the utterance level. Extensive experiments on public datasets are conducted to verify the performance of ALMV under various heterogeneous noise scenarios. Experimental results demonstrate that ALMV substantially improves the robustness and performance of multimodal sentiment analysis in the presence of diverse data defects.
Improving robustness to feature noise is one of the most challenging problems in multimodal sentiment analysis (MSA). Recent studies have proposed efficient MSA models that consider missing modalities, but they typically focus on specific types of defects, which leads to limited robustness and generalization in real-world scenarios where multiple types of noise coexist. In addition, deep interactions across different modalities are often inadequately modeled, making it difficult to fully capture multimodal emotional semantics. To address these issues, this study proposes a robust MSA method that integrates adversarial learning with a multi-view network, namely ALMV. Specifically, temporal modality feature masking is adopted to simulate noisy data, and noise-original instance pairs are constructed with intact sequences for data augmentation. Secondly, temporal convolutional networks and Transformer encoders are used to extract local and global information from every modality sequence, and a weight-controlled multi-view network is constructed to learn joint multimodal representations from the noise-original instance pairs. Additionally, a novel multi-level adversarial training strategy with semantic reconstruction supervision is introduced to learn unified representations between noisy and complete data at both the modality level and the utterance level. Extensive experiments on public datasets are conducted to verify the performance of ALMV under various heterogeneous noise scenarios. Experimental results demonstrate that ALMV substantially improves the robustness and performance of multimodal sentiment analysis in the presence of diverse data defects.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007629
, CSTR: 32375.14.jos.007629
Abstract:
Data distributions in real-world scenarios commonly exhibit long-tail patterns, causing deep models to suffer from head bias, where performance is biased toward head classes while tail classes are poorly recognized. A simple and effective strategy to alleviate this problem is to balance data distributions by augmenting tail-class samples. Although most methods following this strategy achieve a quantitative rebalancing, the generated samples often suffer from semantic shift and insufficient diversity, which leads to dispersed feature distributions and biased decision boundaries, thereby limiting overall model performance. To address these issues, this study proposes a sample-feature collaborative learning framework for long-tailed recognition, aiming to construct classification models that are insensitive to data distribution imbalance. At the sample level, a “magnitude augmentation” strategy based on the Fourier transform is introduced, in which amplitude shifting is employed to modify the style of augmented samples, while preserving their original semantic information. At the feature level, a “feature collapse” loss inspired by neural collapse theory is proposed to align class prototypes into an equiangular tight frame with maximum separability and to encourage features to converge toward their corresponding class prototypes, achieving maximum inter-class separation. By jointly addressing head bias from both the sample and feature perspectives, the proposed framework enhances intra-class compactness and calibrates decision boundaries. Experimental results across multiple benchmark datasets demonstrate that the proposed method significantly improves long-tailed recognition performance.
Data distributions in real-world scenarios commonly exhibit long-tail patterns, causing deep models to suffer from head bias, where performance is biased toward head classes while tail classes are poorly recognized. A simple and effective strategy to alleviate this problem is to balance data distributions by augmenting tail-class samples. Although most methods following this strategy achieve a quantitative rebalancing, the generated samples often suffer from semantic shift and insufficient diversity, which leads to dispersed feature distributions and biased decision boundaries, thereby limiting overall model performance. To address these issues, this study proposes a sample-feature collaborative learning framework for long-tailed recognition, aiming to construct classification models that are insensitive to data distribution imbalance. At the sample level, a “magnitude augmentation” strategy based on the Fourier transform is introduced, in which amplitude shifting is employed to modify the style of augmented samples, while preserving their original semantic information. At the feature level, a “feature collapse” loss inspired by neural collapse theory is proposed to align class prototypes into an equiangular tight frame with maximum separability and to encourage features to converge toward their corresponding class prototypes, achieving maximum inter-class separation. By jointly addressing head bias from both the sample and feature perspectives, the proposed framework enhances intra-class compactness and calibrates decision boundaries. Experimental results across multiple benchmark datasets demonstrate that the proposed method significantly improves long-tailed recognition performance.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007656
, CSTR: 32375.14.jos.007656
Abstract:
Although large language models (LLMs) provide a new exploration path in the field of automated algorithmic theorem proving, existing methods fail to sufficiently improve the capability of automated theorem provers in proving intermediate propositions. Currently, labeled data for formal theorem proving in linear data structure algorithms is very scarce, which presents significant challenges when large language models are used for prediction. To address this issue, this study proposes a novel lemma-library-driven proof method, LDPM, for automated theorem proving of linear data structure algorithms, which integrates large language models with formalization techniques. The method constructs a lemma library composed of newly proven lemmas. By retrieving relevant lemmas from this library, the capability of the automated theorem prover in proving intermediate propositions can be effectively enhanced. The construction of the lemma library follows a three-stage progressive strategy. First, a joint request mechanism is adopted to generate multiple candidate lemma statements, thus enriching the library resources. Second, an iterative proof strategy is employed to prove the candidate lemmas one by one, gradually improving the reliability of the lemma library. Finally, a lemma reflection mechanism is introduced, leveraging LLMs to revise and optimize lemmas, achieving the dynamic expansion and continuous refinement of the lemma library. Through this method, the performance of theorem proving for linear data structure algorithms is significantly improved. Compared with the state-of-the-art methods, the success rate of theorem proving is increased from 44.00% to 82.00%. Ablation experiments further show that the lemma library plays a key role in improving the capability of automated theorem provers in proving intermediate propositions, increasing the success rate of theorem proving for linear data structure algorithms by 38.00%.
Although large language models (LLMs) provide a new exploration path in the field of automated algorithmic theorem proving, existing methods fail to sufficiently improve the capability of automated theorem provers in proving intermediate propositions. Currently, labeled data for formal theorem proving in linear data structure algorithms is very scarce, which presents significant challenges when large language models are used for prediction. To address this issue, this study proposes a novel lemma-library-driven proof method, LDPM, for automated theorem proving of linear data structure algorithms, which integrates large language models with formalization techniques. The method constructs a lemma library composed of newly proven lemmas. By retrieving relevant lemmas from this library, the capability of the automated theorem prover in proving intermediate propositions can be effectively enhanced. The construction of the lemma library follows a three-stage progressive strategy. First, a joint request mechanism is adopted to generate multiple candidate lemma statements, thus enriching the library resources. Second, an iterative proof strategy is employed to prove the candidate lemmas one by one, gradually improving the reliability of the lemma library. Finally, a lemma reflection mechanism is introduced, leveraging LLMs to revise and optimize lemmas, achieving the dynamic expansion and continuous refinement of the lemma library. Through this method, the performance of theorem proving for linear data structure algorithms is significantly improved. Compared with the state-of-the-art methods, the success rate of theorem proving is increased from 44.00% to 82.00%. Ablation experiments further show that the lemma library plays a key role in improving the capability of automated theorem provers in proving intermediate propositions, increasing the success rate of theorem proving for linear data structure algorithms by 38.00%.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007687
, CSTR: 32375.14.jos.007687
Abstract:
With the continuous growth of spatio-temporal data, efficiently managing such data and extracting valuable information from it has become a crucial research topic. On one hand, the core of spatio-temporal data management (STDM) lies in the efficient storage, indexing, and querying of spatio-temporal data. However, traditional database technologies struggle to effectively handle characteristics of spatio-temporal data such as high spatio-temporal dynamics. Artificial intelligence (AI) techniques can effectively capture information such as the distribution characteristics of spatio-temporal data and query workloads, thus making spatio-temporal data management systems more intelligent. On the other hand, spatio-temporal AI (STAI) is devoted to applying techniques such as machine learning, deep learning, and reinforcement learning to spatio-temporal data analysis. STAI enables the automatic identification of patterns, trends, and associations in data and supports tasks such as spatio-temporal prediction, classification, clustering, and anomaly detection. However, across the entire process from data acquisition and model training to practical application, STAI faces challenges such as data heterogeneity, preparation complexity, and high barriers to use. These issues can be effectively mitigated by spatio-temporal data management technologies. Focusing on these topics, extensive research has been conducted in both industry and academia. This study first proposes a taxonomy of spatio-temporal data, categorizing it into independent data and associated data. Then, research progress on the synergy between STDM and STAI is systematically reviewed, and the research background and key techniques are summarized. Finally, commonly used datasets in this field are cataloged, representative applications are introduced, the main challenges are discussed, and future research directions are outlined.
With the continuous growth of spatio-temporal data, efficiently managing such data and extracting valuable information from it has become a crucial research topic. On one hand, the core of spatio-temporal data management (STDM) lies in the efficient storage, indexing, and querying of spatio-temporal data. However, traditional database technologies struggle to effectively handle characteristics of spatio-temporal data such as high spatio-temporal dynamics. Artificial intelligence (AI) techniques can effectively capture information such as the distribution characteristics of spatio-temporal data and query workloads, thus making spatio-temporal data management systems more intelligent. On the other hand, spatio-temporal AI (STAI) is devoted to applying techniques such as machine learning, deep learning, and reinforcement learning to spatio-temporal data analysis. STAI enables the automatic identification of patterns, trends, and associations in data and supports tasks such as spatio-temporal prediction, classification, clustering, and anomaly detection. However, across the entire process from data acquisition and model training to practical application, STAI faces challenges such as data heterogeneity, preparation complexity, and high barriers to use. These issues can be effectively mitigated by spatio-temporal data management technologies. Focusing on these topics, extensive research has been conducted in both industry and academia. This study first proposes a taxonomy of spatio-temporal data, categorizing it into independent data and associated data. Then, research progress on the synergy between STDM and STAI is systematically reviewed, and the research background and key techniques are summarized. Finally, commonly used datasets in this field are cataloged, representative applications are introduced, the main challenges are discussed, and future research directions are outlined.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007688
, CSTR: 32375.14.jos.007688
Abstract:
Although deep neural networks (DNNs) have been widely used in various fields, they exhibit significant vulnerability in adversarial environments. In recent years, a specialized form of attack has emerged, called backdoor attacks or Trojan attacks. Attackers intentionally implant backdoors in DNNs, causing backdoored models to make correct predictions on normal inputs but produce incorrect predictions on inputs containing triggers. To address the threat of backdoor attacks, researchers have proposed various detection methods. However, these methods rely on strict assumptions, such as white-box access to the target model or knowledge of trigger patterns, and require a large amount of detection data and detection time, which limits their applicability in real-world scenarios. Moreover, existing methods have insufficient research on identifying target classes in backdoor attacks and fail to effectively reveal the critical role of target classes in attacks and their impact on real-world tasks. This study proposes an efficient black-box backdoor detection method (EBLD). With only a small amount of clean data, transfer learning is leveraged to train a small set of benign models and backdoored models. Subsequently, a large number of models are rapidly generated through model mutation techniques. Finally, based on these models, multiple classifiers are trained to jointly determine whether the target model is a backdoored model. In the absence of detection data, this study proposes a “feature adaptation” strategy, which leverages benign models and backdoored models generated on other tasks to train binary classifiers capable of detecting new target models. This strategy overcomes the reliance on detection samples and fully utilizes existing benign and backdoored model resources, thus improving detection efficiency. In addition, a binary classifier fine-tuning method based on transfer learning is designed. By loading the weights of existing binary classifiers to train new binary classifiers, backdoor detection time can be further reduced. Finally, an optimal query set library is constructed based on the optimal query sets obtained during binary classifier training. By analyzing the output class distribution of backdoored models on the optimal query sets, the target class of the backdoor attack is quickly identified. Target class identification not only helps to more precisely locate the attack scope but also provides clear guidance for backdoor defense and repair. Experimental results demonstrate that the proposed method significantly outperforms existing methods in detection accuracy and achieves a high level of detection efficiency.
Although deep neural networks (DNNs) have been widely used in various fields, they exhibit significant vulnerability in adversarial environments. In recent years, a specialized form of attack has emerged, called backdoor attacks or Trojan attacks. Attackers intentionally implant backdoors in DNNs, causing backdoored models to make correct predictions on normal inputs but produce incorrect predictions on inputs containing triggers. To address the threat of backdoor attacks, researchers have proposed various detection methods. However, these methods rely on strict assumptions, such as white-box access to the target model or knowledge of trigger patterns, and require a large amount of detection data and detection time, which limits their applicability in real-world scenarios. Moreover, existing methods have insufficient research on identifying target classes in backdoor attacks and fail to effectively reveal the critical role of target classes in attacks and their impact on real-world tasks. This study proposes an efficient black-box backdoor detection method (EBLD). With only a small amount of clean data, transfer learning is leveraged to train a small set of benign models and backdoored models. Subsequently, a large number of models are rapidly generated through model mutation techniques. Finally, based on these models, multiple classifiers are trained to jointly determine whether the target model is a backdoored model. In the absence of detection data, this study proposes a “feature adaptation” strategy, which leverages benign models and backdoored models generated on other tasks to train binary classifiers capable of detecting new target models. This strategy overcomes the reliance on detection samples and fully utilizes existing benign and backdoored model resources, thus improving detection efficiency. In addition, a binary classifier fine-tuning method based on transfer learning is designed. By loading the weights of existing binary classifiers to train new binary classifiers, backdoor detection time can be further reduced. Finally, an optimal query set library is constructed based on the optimal query sets obtained during binary classifier training. By analyzing the output class distribution of backdoored models on the optimal query sets, the target class of the backdoor attack is quickly identified. Target class identification not only helps to more precisely locate the attack scope but also provides clear guidance for backdoor defense and repair. Experimental results demonstrate that the proposed method significantly outperforms existing methods in detection accuracy and achieves a high level of detection efficiency.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007677
, CSTR: 32375.14.jos.007677
Abstract:
With the increasing application of blockchain technology in critical domains such as financial technology and supply chain management, hybrid storage architectures that combine the high performance of traditional databases with the high trustworthiness of blockchains have become a research hotspot. However, existing solutions generally fail to systematically address the architectural incompatibilities between databases and blockchains, which manifest in three core technical challenges: transaction model differences, performance-cost contradictions, and data lifecycle management difficulties. To address these challenges, this study proposes and implements TDSM, a CDC-based database-blockchain synchronization system for achieving efficient and trustworthy data synchronization in heterogeneous environments. First, the study designs a lightweight atomic commit protocol integrated with verifiable revocation proofs and utilizes sparse Merkle trees to generate cryptographic proofs for deletion operations, ensuring the atomic mapping of cross-system transactions without introducing heavyweight two-phase commit, and thus achieving trusted on-chain auditing of the complete data lifecycle. Second, it proposes a resource-aware adaptive batching algorithm, which dynamically adjusts batching strategies by monitoring on-chain transaction costs in real time, effectively resolving the contradiction between the high throughput of databases and the high cost of blockchains. Experimental results demonstrate that TDSM maintains high throughput while significantly reducing synchronization costs. The adaptive batching mechanism achieves substantial resource savings compared to fixed strategies, the revocation proof mechanism remains efficient when processing large numbers of deletion operations, and the atomicity guarantee mechanism ensures data consistency across various failure scenarios.
With the increasing application of blockchain technology in critical domains such as financial technology and supply chain management, hybrid storage architectures that combine the high performance of traditional databases with the high trustworthiness of blockchains have become a research hotspot. However, existing solutions generally fail to systematically address the architectural incompatibilities between databases and blockchains, which manifest in three core technical challenges: transaction model differences, performance-cost contradictions, and data lifecycle management difficulties. To address these challenges, this study proposes and implements TDSM, a CDC-based database-blockchain synchronization system for achieving efficient and trustworthy data synchronization in heterogeneous environments. First, the study designs a lightweight atomic commit protocol integrated with verifiable revocation proofs and utilizes sparse Merkle trees to generate cryptographic proofs for deletion operations, ensuring the atomic mapping of cross-system transactions without introducing heavyweight two-phase commit, and thus achieving trusted on-chain auditing of the complete data lifecycle. Second, it proposes a resource-aware adaptive batching algorithm, which dynamically adjusts batching strategies by monitoring on-chain transaction costs in real time, effectively resolving the contradiction between the high throughput of databases and the high cost of blockchains. Experimental results demonstrate that TDSM maintains high throughput while significantly reducing synchronization costs. The adaptive batching mechanism achieves substantial resource savings compared to fixed strategies, the revocation proof mechanism remains efficient when processing large numbers of deletion operations, and the atomicity guarantee mechanism ensures data consistency across various failure scenarios.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007676
, CSTR: 32375.14.jos.007676
Abstract:
Edge intelligence places models at the network edge, effectively reducing data transmission latency and alleviating the computational load on the central cloud. However, the deployment and continuous updating of models at the edge make deployment efficiency and cost key criteria for evaluating system performance. Although prior work has proposed various strategies to balance model accuracy and deployment costs, in practical settings with volatile service demands and constrained resources, existing methods still struggle to achieve efficient and adaptive deployment and updates. To address this challenge, this study proposes PACE, a low-cost and rapidly adaptive model deployment and update strategy for edge intelligence. PACE aims to achieve efficient responses to dynamic service demands with lower deployment costs, thus enhancing the sustained service capability and performance of edge intelligence systems. Specifically, shared characteristics among models are exploited to perform fine-grained modeling of deployment costs at different stages, and the age of information (AOI) is incorporated to construct a performance evaluation model, thus enabling more cost-effective deployment decisions. On this basis, meta-reinforcement learning is introduced to learn prior knowledge of task variations, and a variational Bayesian network is combined to achieve online optimization, thus enhancing the rapid adaptability of the deployment strategy in dynamic environments. Finally, the effectiveness of PACE is theoretically validated from the perspectives of generalization and convergence. The evaluation is conducted using a synthetic Zipf request distribution. Experimental results show that, in scenarios with frequently changing tasks, PACE achieves efficient adaptive deployment, significantly reducing model deployment costs while maintaining acceptable model accuracy and system response efficiency.
Edge intelligence places models at the network edge, effectively reducing data transmission latency and alleviating the computational load on the central cloud. However, the deployment and continuous updating of models at the edge make deployment efficiency and cost key criteria for evaluating system performance. Although prior work has proposed various strategies to balance model accuracy and deployment costs, in practical settings with volatile service demands and constrained resources, existing methods still struggle to achieve efficient and adaptive deployment and updates. To address this challenge, this study proposes PACE, a low-cost and rapidly adaptive model deployment and update strategy for edge intelligence. PACE aims to achieve efficient responses to dynamic service demands with lower deployment costs, thus enhancing the sustained service capability and performance of edge intelligence systems. Specifically, shared characteristics among models are exploited to perform fine-grained modeling of deployment costs at different stages, and the age of information (AOI) is incorporated to construct a performance evaluation model, thus enabling more cost-effective deployment decisions. On this basis, meta-reinforcement learning is introduced to learn prior knowledge of task variations, and a variational Bayesian network is combined to achieve online optimization, thus enhancing the rapid adaptability of the deployment strategy in dynamic environments. Finally, the effectiveness of PACE is theoretically validated from the perspectives of generalization and convergence. The evaluation is conducted using a synthetic Zipf request distribution. Experimental results show that, in scenarios with frequently changing tasks, PACE achieves efficient adaptive deployment, significantly reducing model deployment costs while maintaining acceptable model accuracy and system response efficiency.
Available online: July 08, 2026
, DOI: 10.13328/j.cnki.jos.007673
, CSTR: 32375.14.jos.007673
Abstract:
Artificial intelligence systems rely on large-scale, open training data collection, which creates opportunities for data poisoning attacks. Attackers can manipulate training data by infiltrating the data supply chain, thus compromising the availability of deep learning models. Compared to dirty-label data poisoning attacks based on label flipping, clean-label attacks are more stealthy but face greater challenges in implementation. On the one hand, precise manipulation of sample features is required rather than direct alteration of labels. On the other hand, it is necessary to ensure that poisoned samples remain highly visually consistent with the original samples to evade detection. Although existing clean-label attacks have achieved some success, they are often difficult to implement in practical scenarios due to loose assumptions, and there remains room for improvement in terms of stealthiness and computational efficiency. To address these issues, this study proposes a clean-label data poisoning attack scheme for deep learning based on data reconstruction, termed CPDR. In this scheme, the loss function of the Attention U-Net model, originally designed for image segmentation, is reconstructed and transformed into a multi-task learning reconstruction model that incorporates reconstruction loss, attack effectiveness loss, and model update consistency loss. Through data reconstruction, stealthy poisoned samples are generated, and effective poisoning attacks are achieved under a black-box assumption. Experiments conducted in both centralized and federated learning scenarios, across two datasets and four target models, demonstrate that CPDR reduces the model’s classification accuracy on target classes with minimal impact on the performance of the main task. In centralized learning, CPDR achieves an additional reduction of 2.33 and 2.42 percentage points in target class accuracy compared to VagueGAN and Adversarial Poison, respectively. In federated learning, this advantage further increases to 4.44 and 4.10 percentage points, respectively. Moreover, the poisoned samples generated by CPDR exhibit minimal visual differences from the original samples, making them difficult to distinguish and detect, thus achieving a good balance between stealthiness and attack effectiveness.
Artificial intelligence systems rely on large-scale, open training data collection, which creates opportunities for data poisoning attacks. Attackers can manipulate training data by infiltrating the data supply chain, thus compromising the availability of deep learning models. Compared to dirty-label data poisoning attacks based on label flipping, clean-label attacks are more stealthy but face greater challenges in implementation. On the one hand, precise manipulation of sample features is required rather than direct alteration of labels. On the other hand, it is necessary to ensure that poisoned samples remain highly visually consistent with the original samples to evade detection. Although existing clean-label attacks have achieved some success, they are often difficult to implement in practical scenarios due to loose assumptions, and there remains room for improvement in terms of stealthiness and computational efficiency. To address these issues, this study proposes a clean-label data poisoning attack scheme for deep learning based on data reconstruction, termed CPDR. In this scheme, the loss function of the Attention U-Net model, originally designed for image segmentation, is reconstructed and transformed into a multi-task learning reconstruction model that incorporates reconstruction loss, attack effectiveness loss, and model update consistency loss. Through data reconstruction, stealthy poisoned samples are generated, and effective poisoning attacks are achieved under a black-box assumption. Experiments conducted in both centralized and federated learning scenarios, across two datasets and four target models, demonstrate that CPDR reduces the model’s classification accuracy on target classes with minimal impact on the performance of the main task. In centralized learning, CPDR achieves an additional reduction of 2.33 and 2.42 percentage points in target class accuracy compared to VagueGAN and Adversarial Poison, respectively. In federated learning, this advantage further increases to 4.44 and 4.10 percentage points, respectively. Moreover, the poisoned samples generated by CPDR exhibit minimal visual differences from the original samples, making them difficult to distinguish and detect, thus achieving a good balance between stealthiness and attack effectiveness.
Available online: July 01, 2026
, DOI: 10.13328/j.cnki.jos.007675
, CSTR: 32375.14.jos.007675
Abstract:
Prompt injection attacks have become an increasingly serious security threat to large language models (LLMs) and are ranked first in the OWASP Top 10 for LLM applications. Such attacks induce LLMs to generate misleading outputs by embedding malicious content in prompts, thus posing serious security risks such as information leakage and system abuse. However, current evaluation metrics for attack strategies against such attacks only include the attack success rate, making it difficult to comprehensively assess attack effectiveness. In terms of defense strategies, existing methods can typically be used for either prevention or detection, but it is difficult to achieve both simultaneously. This study presents an in-depth analysis of the characteristics of prompt injection attacks and defenses, as well as their impact on LLM outputs, and proposes an integrated attack-defense framework termed Int-AD. The proposed framework includes a prompt injection attack method, control-output attack (CoA), which employs sentiment enhancement and output control, and a defense strategy, universal defense (UnD), which possesses both prevention and detection capabilities. In terms of attacks, two new evaluation metrics are defined: attack interference rate (AIR) and attack misdirection rate (AMR), which are used to conduct precise evaluations under different attack outcomes, thus improving the evaluation metrics. By designing prompts with sentiment enhancement and output control, AIR and AMR are combined with the existing METEOR evaluation metric to assess the attack efficiency of CoA. On the defense side, UnD possesses both prevention and detection capabilities, which improves the defense success rate and enables the detection of the impact of prompt injection attacks. Experiments are conducted on three mainstream large models, Qwen, Llama, and DeepSeek, using the squad and web_questions public datasets. The average METEOR score of the CoA attack method is 0.050, which is 42.5% lower than that of the current best method, Combine. The average AIR and AMR reach 0.97 and 0.45, respectively, representing improvements of 6.59% and 28.57% over Combine and indicating that CoA possesses stronger attack capability. The average defense success rate of UnD is 80.6%, which is 9.07% higher than that of the current best method, Sandwich. Its average knownanswer matching rate (KMR) is 0.921, which is 58.52% higher than that of the Knownanswer method, indicating that UnD possesses stronger defense capability.
Prompt injection attacks have become an increasingly serious security threat to large language models (LLMs) and are ranked first in the OWASP Top 10 for LLM applications. Such attacks induce LLMs to generate misleading outputs by embedding malicious content in prompts, thus posing serious security risks such as information leakage and system abuse. However, current evaluation metrics for attack strategies against such attacks only include the attack success rate, making it difficult to comprehensively assess attack effectiveness. In terms of defense strategies, existing methods can typically be used for either prevention or detection, but it is difficult to achieve both simultaneously. This study presents an in-depth analysis of the characteristics of prompt injection attacks and defenses, as well as their impact on LLM outputs, and proposes an integrated attack-defense framework termed Int-AD. The proposed framework includes a prompt injection attack method, control-output attack (CoA), which employs sentiment enhancement and output control, and a defense strategy, universal defense (UnD), which possesses both prevention and detection capabilities. In terms of attacks, two new evaluation metrics are defined: attack interference rate (AIR) and attack misdirection rate (AMR), which are used to conduct precise evaluations under different attack outcomes, thus improving the evaluation metrics. By designing prompts with sentiment enhancement and output control, AIR and AMR are combined with the existing METEOR evaluation metric to assess the attack efficiency of CoA. On the defense side, UnD possesses both prevention and detection capabilities, which improves the defense success rate and enables the detection of the impact of prompt injection attacks. Experiments are conducted on three mainstream large models, Qwen, Llama, and DeepSeek, using the squad and web_questions public datasets. The average METEOR score of the CoA attack method is 0.050, which is 42.5% lower than that of the current best method, Combine. The average AIR and AMR reach 0.97 and 0.45, respectively, representing improvements of 6.59% and 28.57% over Combine and indicating that CoA possesses stronger attack capability. The average defense success rate of UnD is 80.6%, which is 9.07% higher than that of the current best method, Sandwich. Its average knownanswer matching rate (KMR) is 0.921, which is 58.52% higher than that of the Knownanswer method, indicating that UnD possesses stronger defense capability.
Available online: July 01, 2026
, DOI: 10.13328/j.cnki.jos.007678
, CSTR: 32375.14.jos.007678
Abstract:
Heuristic data augmentation effectively mitigates data sparsity in sequential recommendation by perturbing the original user behavior sequences and generating new ones. Compared with augmentation methods that rely on external information, complex augmentation rules, or learnable generation modules, heuristic methods perform augmentation through simple operations based on prior human knowledge, rules, or intuition, offering high efficiency and generalizability. However, existing heuristic data augmentation methods typically face two challenges. First, the augmented data often exhibits semantic bias or loss of important user behaviors compared with the original data, leading models to learn inaccurate or even erroneous user preferences. Second, existing approaches focus on augmenting sparse input behaviors, yet the model training remains constrained by the sparsity and mismatch in labeled behaviors, limiting improvements in recommendation performance. To address this issue, this study proposes a framework for repairing and enriching heuristic data augmentation in sequential recommendation (REDRec). Specifically, this study first validates the semantic deviation in sequence preferences caused by heuristic recommendation methods through empirical studies and demonstrates that a simple Jensen-Shannon divergence filtering strategy can mitigate this issue. On this basis, a repair mechanism that is sensitive to the degree of semantic shift is designed. REDRec quantifies the semantic deviation of augmented sequences using Jensen-Shannon divergence and performs an embedding representation interpolation based on differentiated beta distributions for samples with varying levels of deviation to achieve semantic repair. To enrich the labels of augmented data, REDRec performs perturbation-based multi-round inference on repaired samples and generates new soft labels by adaptively aggregating these inference results. The new soft labels are added as additional data during model training. Experiments on multiple representative backbone networks and real-world datasets validate the effectiveness and generalization of REDRec.
Heuristic data augmentation effectively mitigates data sparsity in sequential recommendation by perturbing the original user behavior sequences and generating new ones. Compared with augmentation methods that rely on external information, complex augmentation rules, or learnable generation modules, heuristic methods perform augmentation through simple operations based on prior human knowledge, rules, or intuition, offering high efficiency and generalizability. However, existing heuristic data augmentation methods typically face two challenges. First, the augmented data often exhibits semantic bias or loss of important user behaviors compared with the original data, leading models to learn inaccurate or even erroneous user preferences. Second, existing approaches focus on augmenting sparse input behaviors, yet the model training remains constrained by the sparsity and mismatch in labeled behaviors, limiting improvements in recommendation performance. To address this issue, this study proposes a framework for repairing and enriching heuristic data augmentation in sequential recommendation (REDRec). Specifically, this study first validates the semantic deviation in sequence preferences caused by heuristic recommendation methods through empirical studies and demonstrates that a simple Jensen-Shannon divergence filtering strategy can mitigate this issue. On this basis, a repair mechanism that is sensitive to the degree of semantic shift is designed. REDRec quantifies the semantic deviation of augmented sequences using Jensen-Shannon divergence and performs an embedding representation interpolation based on differentiated beta distributions for samples with varying levels of deviation to achieve semantic repair. To enrich the labels of augmented data, REDRec performs perturbation-based multi-round inference on repaired samples and generates new soft labels by adaptively aggregating these inference results. The new soft labels are added as additional data during model training. Experiments on multiple representative backbone networks and real-world datasets validate the effectiveness and generalization of REDRec.
Available online: July 01, 2026
, DOI: 10.13328/j.cnki.jos.007685
, CSTR: 32375.14.jos.007685
Abstract:
With the rapid adoption of the Solana blockchain, prior research has examined its security vulnerabilities, ecosystem, and commercial applications. However, concurrency defects in Solana on-chain programs have not received sufficient attention. This study identifies two special types of concurrency defects in Solana on-chain programs: transaction-order dependency defects and clock-control dependency defects. These defects allow malicious nodes in the blockchain network to legitimately reorder concurrent transactions or delay specific transactions until a specific time, thus affecting the execution outcomes of these concurrent transactions. As a result, concurrent transactions in Solana on-chain programs containing these defects may be vulnerable to adversarial control. To address this problem, this study defines transaction-order dependency defects and clock-control dependency defects in Solana on-chain programs and proposes a detection method based on fuzz testing and symbolic execution. The method includes detection criteria for both defect types, as well as a ledger simulation method and a transaction generation method tailored to the characteristics of the Solana blockchain. Based on this method, a prototype tool, SolRaceFuzz, is implemented, and experiments are conducted on three datasets to evaluate defect discovery capability, efficiency, and exploration strategy, verifying the effectiveness of the proposed method.
With the rapid adoption of the Solana blockchain, prior research has examined its security vulnerabilities, ecosystem, and commercial applications. However, concurrency defects in Solana on-chain programs have not received sufficient attention. This study identifies two special types of concurrency defects in Solana on-chain programs: transaction-order dependency defects and clock-control dependency defects. These defects allow malicious nodes in the blockchain network to legitimately reorder concurrent transactions or delay specific transactions until a specific time, thus affecting the execution outcomes of these concurrent transactions. As a result, concurrent transactions in Solana on-chain programs containing these defects may be vulnerable to adversarial control. To address this problem, this study defines transaction-order dependency defects and clock-control dependency defects in Solana on-chain programs and proposes a detection method based on fuzz testing and symbolic execution. The method includes detection criteria for both defect types, as well as a ledger simulation method and a transaction generation method tailored to the characteristics of the Solana blockchain. Based on this method, a prototype tool, SolRaceFuzz, is implemented, and experiments are conducted on three datasets to evaluate defect discovery capability, efficiency, and exploration strategy, verifying the effectiveness of the proposed method.
Available online: July 01, 2026
, DOI: 10.13328/j.cnki.jos.007686
, CSTR: 32375.14.jos.007686
Abstract:
In recent years, deep learning has provided a new research path for block cipher security analysis. However, for large-state block ciphers, the high input dimensionality of neural distinguishers and the rapid expansion of the key search space severely limit the efficiency and stability of deep learning-assisted attacks. To address these issues, this study proposes DLAKRS, a deep learning-assisted key recovery scheme for large-state SIMECK block ciphers. In the training phase, the scheme introduces a gradient-based dynamic sensitive bit selection strategy to effectively compress the input dimension of the neural distinguisher while ensuring distinguishing performance. In the key recovery phase, a statistical score-guided search method is constructed based on the wrong-key response distribution to alleviate score fluctuations in the candidate key screening process. Based on the above scheme, this study constructs a multi-stage deep learning-assisted key recovery attack scheme for the SIMECK series of large-state lightweight block ciphers. The experimental results show that, when the input dimension is reduced by 45.8%–62.5%, the accuracy loss of the distinguisher is controlled within 10%, and the computational cost is significantly reduced compared with traditional multi-stage attack methods. In 16-round attacks on SIMECK48/96 and 18-round attacks on SIMECK64/128, the proposed scheme shows stable key recovery performance. These results show that the proposed scheme provides an effective way for deep learning-assisted key recovery of large-state SIMECK block ciphers.
In recent years, deep learning has provided a new research path for block cipher security analysis. However, for large-state block ciphers, the high input dimensionality of neural distinguishers and the rapid expansion of the key search space severely limit the efficiency and stability of deep learning-assisted attacks. To address these issues, this study proposes DLAKRS, a deep learning-assisted key recovery scheme for large-state SIMECK block ciphers. In the training phase, the scheme introduces a gradient-based dynamic sensitive bit selection strategy to effectively compress the input dimension of the neural distinguisher while ensuring distinguishing performance. In the key recovery phase, a statistical score-guided search method is constructed based on the wrong-key response distribution to alleviate score fluctuations in the candidate key screening process. Based on the above scheme, this study constructs a multi-stage deep learning-assisted key recovery attack scheme for the SIMECK series of large-state lightweight block ciphers. The experimental results show that, when the input dimension is reduced by 45.8%–62.5%, the accuracy loss of the distinguisher is controlled within 10%, and the computational cost is significantly reduced compared with traditional multi-stage attack methods. In 16-round attacks on SIMECK48/96 and 18-round attacks on SIMECK64/128, the proposed scheme shows stable key recovery performance. These results show that the proposed scheme provides an effective way for deep learning-assisted key recovery of large-state SIMECK block ciphers.
Available online: June 24, 2026
, DOI: 10.13328/j.cnki.jos.007684
Abstract:
With the continuous advancement of deep learning and Transformer architectures, natural language processing tasks based on large language models have achieved remarkable progress. However, issues such as hallucinations, information lag, and lack of domain-specific knowledge persist. To address these issues, retrieval-augmented generation techniques introduce retrieval mechanisms based on external knowledge repositories, enabling models not only to use existing knowledge but also to retrieve external data in real time, thus improving the accuracy, timeliness, and adaptability of generated content. This study examines how retrieval-augmented generation enhances the generative capabilities of large language models. First, from the perspectives of retrieval augmentation and augmented generation, it systematically classifies retrieval methods according to different retrieval objectives and multiple technical pathways for augmented generation. Subsequently, the principles, implementation methods, and applications of retrieval-augmented generation technologies across various domains are analyzed and summarized. Finally, the current applications, challenges, and future prospects of retrieval-augmented generation across multiple fields are examined.
With the continuous advancement of deep learning and Transformer architectures, natural language processing tasks based on large language models have achieved remarkable progress. However, issues such as hallucinations, information lag, and lack of domain-specific knowledge persist. To address these issues, retrieval-augmented generation techniques introduce retrieval mechanisms based on external knowledge repositories, enabling models not only to use existing knowledge but also to retrieve external data in real time, thus improving the accuracy, timeliness, and adaptability of generated content. This study examines how retrieval-augmented generation enhances the generative capabilities of large language models. First, from the perspectives of retrieval augmentation and augmented generation, it systematically classifies retrieval methods according to different retrieval objectives and multiple technical pathways for augmented generation. Subsequently, the principles, implementation methods, and applications of retrieval-augmented generation technologies across various domains are analyzed and summarized. Finally, the current applications, challenges, and future prospects of retrieval-augmented generation across multiple fields are examined.
Available online: June 24, 2026
, DOI: 10.13328/j.cnki.jos.007627
Abstract:
Knowledge graphs effectively organize and exploit diverse real-world knowledge, alleviating limitations of purely data-driven methods and large language models, such as limited interpretability and hallucination, and serve as one of the core technologies for cognitive intelligent systems with understanding and reasoning capabilities. Knowledge graph reasoning aims to infer new knowledge from existing knowledge and is widely applied in various fields, including industrial and agricultural production, national defense and security, and daily life. Early studies on knowledge graph reasoning primarily focus on single tasks, while recent research increasingly shifts attention to more complex tasks. These complex tasks include temporal multi-step knowledge graph reasoning, few-shot multi-step knowledge graph reasoning, few-shot temporal knowledge graph reasoning, few-shot multi-modal knowledge graph reasoning, multi-modal inductive knowledge graph reasoning, and temporal inductive knowledge graph reasoning. Accordingly, this study first describes the characteristics of knowledge graph reasoning for single tasks and complex tasks. Subsequently, several representative knowledge graph reasoning techniques for single tasks are introduced. Then, existing representative knowledge graph reasoning methods for complex tasks and conduct a comparative analysis of these methods are systematically summarized and comparatively analyzed. Finally, the major remaining challenges in knowledge graph reasoning are discussed, and promising future research directions are outlined.
Knowledge graphs effectively organize and exploit diverse real-world knowledge, alleviating limitations of purely data-driven methods and large language models, such as limited interpretability and hallucination, and serve as one of the core technologies for cognitive intelligent systems with understanding and reasoning capabilities. Knowledge graph reasoning aims to infer new knowledge from existing knowledge and is widely applied in various fields, including industrial and agricultural production, national defense and security, and daily life. Early studies on knowledge graph reasoning primarily focus on single tasks, while recent research increasingly shifts attention to more complex tasks. These complex tasks include temporal multi-step knowledge graph reasoning, few-shot multi-step knowledge graph reasoning, few-shot temporal knowledge graph reasoning, few-shot multi-modal knowledge graph reasoning, multi-modal inductive knowledge graph reasoning, and temporal inductive knowledge graph reasoning. Accordingly, this study first describes the characteristics of knowledge graph reasoning for single tasks and complex tasks. Subsequently, several representative knowledge graph reasoning techniques for single tasks are introduced. Then, existing representative knowledge graph reasoning methods for complex tasks and conduct a comparative analysis of these methods are systematically summarized and comparatively analyzed. Finally, the major remaining challenges in knowledge graph reasoning are discussed, and promising future research directions are outlined.
Available online: June 24, 2026
, DOI: 10.13328/j.cnki.jos.007672
Abstract:
Identifying influential nodes is crucial in social network analysis. Existing methods often neglect the tendencies of local opinion leaders, resulting in overlapping influence regions among seed nodes. Furthermore, in dynamic scenarios with non-uniform influence propagation intensity, approaches based on vanilla graph neural networks (GNNs) struggle to effectively model and aggregate node influence characteristics during message passing. Current techniques also fail to adequately address the multilayer nature of social networks and node heterogeneity. To address these issues, this study proposes Inf-MDE, a novel multilayer influence maximization method leveraging differentiated graph embeddings, which models social relationships using a multilayer network structure. The model further extracts latent influence propagation subgraphs of nodes to eliminate the representation bias between node embeddings and propagation dynamics. In addition, Inf-MDE incorporates an adaptive local influence aggregation mechanism within its GNN design. This mechanism dynamically adjusts the aggregation strategy of influence features during message passing based on local context and influence intensity, thus effectively capturing inter-layer propagation heterogeneity and intra-layer diffusion dynamics. Extensive experiments conducted on four different multilayer social network datasets demonstrate that Inf-MDE significantly outperforms existing baseline methods. The source code of this study is available at https://github.com/lyao972/Inf-MDE.
Identifying influential nodes is crucial in social network analysis. Existing methods often neglect the tendencies of local opinion leaders, resulting in overlapping influence regions among seed nodes. Furthermore, in dynamic scenarios with non-uniform influence propagation intensity, approaches based on vanilla graph neural networks (GNNs) struggle to effectively model and aggregate node influence characteristics during message passing. Current techniques also fail to adequately address the multilayer nature of social networks and node heterogeneity. To address these issues, this study proposes Inf-MDE, a novel multilayer influence maximization method leveraging differentiated graph embeddings, which models social relationships using a multilayer network structure. The model further extracts latent influence propagation subgraphs of nodes to eliminate the representation bias between node embeddings and propagation dynamics. In addition, Inf-MDE incorporates an adaptive local influence aggregation mechanism within its GNN design. This mechanism dynamically adjusts the aggregation strategy of influence features during message passing based on local context and influence intensity, thus effectively capturing inter-layer propagation heterogeneity and intra-layer diffusion dynamics. Extensive experiments conducted on four different multilayer social network datasets demonstrate that Inf-MDE significantly outperforms existing baseline methods. The source code of this study is available at https://github.com/lyao972/Inf-MDE.
Available online: June 10, 2026
, DOI: 10.13328/j.cnki.jos.007657
, CSTR: 32375.14.jos.007657
Abstract:
With the rapid advancement of modern hardware, software and artificial intelligence technologies, intelligent models represented by neural networks have been widely applied across various industries. However, these models also reveal significant safety problems, such as the robustness-related problem of adversarial examples. Therefore, ensuring and detecting the robustness and safety of neural networks through formal verification methods is of paramount importance. However, most existing research on neural network verification primarily targets feedforward neural networks with ReLU activation functions. In contrast, verification efforts for recurrent neural networks (RNN), which feature more complex structures and nonlinear activation functions, are relatively limited. To address this issue, this study proposes a verification method for RNNs based on linear relaxation and dual optimization. This method provides rigorous robustness guarantees by computing the robustness radius of these networks. First, the RNN verification problem is formulated as an optimization problem relating the network’s output to its input through linear relaxation. Taking into account the hierarchical structure of the network, the optimization problem is decomposed using Lagrangian decomposition into subproblems, each involving only two adjacent neuron layers, with equality constraints linking these subproblems. Next, the dual problem is formulated and solved using a gradient ascent algorithm to obtain approximate upper and lower bounds for the output neurons, with the validity of the solution ensured by the properties of dual optimization. Finally, binary search is applied to compute the network’s robustness radius with respect to input perturbations. By conducting experimental comparisons with existing verification methods on vanilla RNNs of different structures and more complex long short-term memory (LSTM) networks, results demonstrate that the proposed method achieves a 17.7% to 30.2% improvement in the robustness radius for vanilla RNNs and a 5.2% to 10.2% improvement for LSTM networks.
With the rapid advancement of modern hardware, software and artificial intelligence technologies, intelligent models represented by neural networks have been widely applied across various industries. However, these models also reveal significant safety problems, such as the robustness-related problem of adversarial examples. Therefore, ensuring and detecting the robustness and safety of neural networks through formal verification methods is of paramount importance. However, most existing research on neural network verification primarily targets feedforward neural networks with ReLU activation functions. In contrast, verification efforts for recurrent neural networks (RNN), which feature more complex structures and nonlinear activation functions, are relatively limited. To address this issue, this study proposes a verification method for RNNs based on linear relaxation and dual optimization. This method provides rigorous robustness guarantees by computing the robustness radius of these networks. First, the RNN verification problem is formulated as an optimization problem relating the network’s output to its input through linear relaxation. Taking into account the hierarchical structure of the network, the optimization problem is decomposed using Lagrangian decomposition into subproblems, each involving only two adjacent neuron layers, with equality constraints linking these subproblems. Next, the dual problem is formulated and solved using a gradient ascent algorithm to obtain approximate upper and lower bounds for the output neurons, with the validity of the solution ensured by the properties of dual optimization. Finally, binary search is applied to compute the network’s robustness radius with respect to input perturbations. By conducting experimental comparisons with existing verification methods on vanilla RNNs of different structures and more complex long short-term memory (LSTM) networks, results demonstrate that the proposed method achieves a 17.7% to 30.2% improvement in the robustness radius for vanilla RNNs and a 5.2% to 10.2% improvement for LSTM networks.
Available online: June 10, 2026
, DOI: 10.13328/j.cnki.jos.007659
, CSTR: 32375.14.jos.007659
Abstract:
A slow query refers to a query whose execution time exceeds the threshold defined by the database system, thus degrading overall performance. Due to limited developer experience and inflexible query interface templates, slow queries frequently occur in applications and are a major cause of database failures. To accelerate slow queries, database administrators often perform manual tuning. However, manual tuning is inefficient for large volumes of data, making research on automatic query rewriting methods for slow queries particularly important. Query hints are tuning directives that explicitly influence optimizer decisions and can improve query performance by affecting strategies such as access paths. As an important research problem in the field of database query rewriting, query hint recommendation has received widespread attention. However, existing machine learning-based hint recommendation methods still face issues such as coarse granularity and reliance on manually specified candidate sets. To address these issues, this study proposes QueryGuide, a fine-grained hint recommendation framework based on a generative model. By exploring the optimal combinations of different fine-grained hints, the execution efficiency of physical plans for rewritten queries is improved. Specifically, on the one hand, the framework explores recommendation strategies based on generative language models, investigates the performance improvement of hints on slow queries in database environments, and uses it as rewards for model fine-tuning, thus generating fine-grained, table-level hints. On the other hand, the fine-grained hints recommended by the generative model are used as exploration directions, and hint combinations are obtained through expansion. The exploration results of these combinations are used for model fine-tuning, enabling the model to learn the impact of different hint combinations on slow query rewriting performance and to identify the optimal hint combination. Experimental results on multiple datasets demonstrate that, compared with other baseline algorithms, the proposed method exhibits better rewriting performance and significantly reduces query execution time.
A slow query refers to a query whose execution time exceeds the threshold defined by the database system, thus degrading overall performance. Due to limited developer experience and inflexible query interface templates, slow queries frequently occur in applications and are a major cause of database failures. To accelerate slow queries, database administrators often perform manual tuning. However, manual tuning is inefficient for large volumes of data, making research on automatic query rewriting methods for slow queries particularly important. Query hints are tuning directives that explicitly influence optimizer decisions and can improve query performance by affecting strategies such as access paths. As an important research problem in the field of database query rewriting, query hint recommendation has received widespread attention. However, existing machine learning-based hint recommendation methods still face issues such as coarse granularity and reliance on manually specified candidate sets. To address these issues, this study proposes QueryGuide, a fine-grained hint recommendation framework based on a generative model. By exploring the optimal combinations of different fine-grained hints, the execution efficiency of physical plans for rewritten queries is improved. Specifically, on the one hand, the framework explores recommendation strategies based on generative language models, investigates the performance improvement of hints on slow queries in database environments, and uses it as rewards for model fine-tuning, thus generating fine-grained, table-level hints. On the other hand, the fine-grained hints recommended by the generative model are used as exploration directions, and hint combinations are obtained through expansion. The exploration results of these combinations are used for model fine-tuning, enabling the model to learn the impact of different hint combinations on slow query rewriting performance and to identify the optimal hint combination. Experimental results on multiple datasets demonstrate that, compared with other baseline algorithms, the proposed method exhibits better rewriting performance and significantly reduces query execution time.
Available online: June 10, 2026
, DOI: 10.13328/j.cnki.jos.007647
, CSTR: 32375.14.jos.007647
Abstract:
Attribute-based encryption (ABE) enables fine-grained access control based on user attributes, but it relies on a centralized authority for key generation, which may cause security risks such as key leakage. Registered-based encryption schemes mitigate this issue using mechanisms such as user registration and public key aggregation. However, existing constructions based on multiple parallel sRABE sub-schemes require frequent public key re-aggregation when users dynamically join or leave, leading to computational and storage overhead that grows linearly with the number of sub-schemes. Therefore, these constructions are difficult to apply in scenarios with frequent user changes. To overcome this limitation, this study proposes a chameleon-hash-based revocable registered attribute-based encryption scheme that achieves efficient user registration and revocation within a single sRABE framework. By introducing a trapdoor-controllable chameleon hash function and allowing the public key aggregator to maintain public key consistency, new users can be registered without system-level public key re-aggregation. Meanwhile, user revocation only requires marking the status of registration slots, thus avoiding the computational overhead caused by the multi-sub-scheme structure. In terms of efficiency, the computational overhead of the new-user registration phase is analyzed, and the results show that the computational complexity of this process depends only on the number of user attributes and is independent of the number of users in the system. Security analysis shows that the proposed scheme is secure against chosen-plaintext attacks under the indistinguishability model.
Attribute-based encryption (ABE) enables fine-grained access control based on user attributes, but it relies on a centralized authority for key generation, which may cause security risks such as key leakage. Registered-based encryption schemes mitigate this issue using mechanisms such as user registration and public key aggregation. However, existing constructions based on multiple parallel sRABE sub-schemes require frequent public key re-aggregation when users dynamically join or leave, leading to computational and storage overhead that grows linearly with the number of sub-schemes. Therefore, these constructions are difficult to apply in scenarios with frequent user changes. To overcome this limitation, this study proposes a chameleon-hash-based revocable registered attribute-based encryption scheme that achieves efficient user registration and revocation within a single sRABE framework. By introducing a trapdoor-controllable chameleon hash function and allowing the public key aggregator to maintain public key consistency, new users can be registered without system-level public key re-aggregation. Meanwhile, user revocation only requires marking the status of registration slots, thus avoiding the computational overhead caused by the multi-sub-scheme structure. In terms of efficiency, the computational overhead of the new-user registration phase is analyzed, and the results show that the computational complexity of this process depends only on the number of user attributes and is independent of the number of users in the system. Security analysis shows that the proposed scheme is secure against chosen-plaintext attacks under the indistinguishability model.
Available online: June 03, 2026
, DOI: 10.13328/j.cnki.jos.007661
Abstract:
Federated Learning enables collaborative model training among multiple clients without exchanging their local private data. However, this paradigm is vulnerable to increasingly sophisticated poisoning attacks. Existing defense mechanisms often exhibit shortcomings in detection efficiency, generalization to diverse attacks, and performance stability under non-independent and identically distributed (non-IID) data. To address these challenges, this study proposes FedProbe, an interpretable federated poisoning attack defense framework guided by proxy models. FedProbe adopts a two-stage mechanism. In the first stage, under proxy model guidance, the framework computes the KL divergence among local models on a trusted server-side dataset to measure their behavioral similarity and determine the optimal number of clusters. After removing outlier updates, each cluster is aggregated into a proxy model, aiming to effectively partition models trained on different data distributions and improve subsequent detection efficiency. In the second stage, an interpretable analysis is conducted. FedProbe leverages the SHAP technique to perform an in-depth analysis of the proxy models, computing key feature attributions and ultimately using cross-category suspicion scores to identify and remove potentially malicious models accurately. Experimental results demonstrate that FedProbe exhibits superior performance and robustness across various benchmark datasets. In benign settings, the convergence time is only 1.11 to 1.21 times that of standard aggregation algorithms, indicating minimal overhead. In terms of security, even in extreme scenarios with up to 40% malicious clients, FedProbe maintains superior convergence when defending against untargeted attacks and backdoor attacks, while keeping the attack success rate below 25%. Moreover, when facing adaptive attacks, it shows significant advantages in both defense effectiveness and task accuracy.
Federated Learning enables collaborative model training among multiple clients without exchanging their local private data. However, this paradigm is vulnerable to increasingly sophisticated poisoning attacks. Existing defense mechanisms often exhibit shortcomings in detection efficiency, generalization to diverse attacks, and performance stability under non-independent and identically distributed (non-IID) data. To address these challenges, this study proposes FedProbe, an interpretable federated poisoning attack defense framework guided by proxy models. FedProbe adopts a two-stage mechanism. In the first stage, under proxy model guidance, the framework computes the KL divergence among local models on a trusted server-side dataset to measure their behavioral similarity and determine the optimal number of clusters. After removing outlier updates, each cluster is aggregated into a proxy model, aiming to effectively partition models trained on different data distributions and improve subsequent detection efficiency. In the second stage, an interpretable analysis is conducted. FedProbe leverages the SHAP technique to perform an in-depth analysis of the proxy models, computing key feature attributions and ultimately using cross-category suspicion scores to identify and remove potentially malicious models accurately. Experimental results demonstrate that FedProbe exhibits superior performance and robustness across various benchmark datasets. In benign settings, the convergence time is only 1.11 to 1.21 times that of standard aggregation algorithms, indicating minimal overhead. In terms of security, even in extreme scenarios with up to 40% malicious clients, FedProbe maintains superior convergence when defending against untargeted attacks and backdoor attacks, while keeping the attack success rate below 25%. Moreover, when facing adaptive attacks, it shows significant advantages in both defense effectiveness and task accuracy.
Available online: June 03, 2026
, DOI: 10.13328/j.cnki.jos.007662
Abstract:
Traditional image steganography achieves covert information transmission by embedding secret information into cover images, playing a crucial role in information security and data communication. However, the information embedding process inevitably alters the cover image, making it more susceptible to detection by steganalysis tools. In contrast, generative image steganography directly synthesizes stego images from secret information using generative models, avoiding modifications to cover images. Nevertheless, existing generative image steganography methods generally suffer from insufficient robustness against various attacks, particularly geometric attacks, often resulting in damage to the hidden information and failure to effectively extract it. To address this issue, this study proposes a generative steganography method guided by human posture (GSHP). The core idea is to map secret information into human posture features, which are then fed into a generative model to produce stego images. During the information extraction stage, a human posture detection algorithm is used to identify the posture in the stego image, thus recovering the hidden information. Owing to the inherent structural stability of human posture, GSHP exhibits good robustness against various attacks. Extensive experiments also fully demonstrate the advantages of GSHP in terms of security and robustness.
Traditional image steganography achieves covert information transmission by embedding secret information into cover images, playing a crucial role in information security and data communication. However, the information embedding process inevitably alters the cover image, making it more susceptible to detection by steganalysis tools. In contrast, generative image steganography directly synthesizes stego images from secret information using generative models, avoiding modifications to cover images. Nevertheless, existing generative image steganography methods generally suffer from insufficient robustness against various attacks, particularly geometric attacks, often resulting in damage to the hidden information and failure to effectively extract it. To address this issue, this study proposes a generative steganography method guided by human posture (GSHP). The core idea is to map secret information into human posture features, which are then fed into a generative model to produce stego images. During the information extraction stage, a human posture detection algorithm is used to identify the posture in the stego image, thus recovering the hidden information. Owing to the inherent structural stability of human posture, GSHP exhibits good robustness against various attacks. Extensive experiments also fully demonstrate the advantages of GSHP in terms of security and robustness.
Available online: June 03, 2026
, DOI: 10.13328/j.cnki.jos.007663
Abstract:
In smart home systems, scenario automation rules can automatically execute a set of device operations based on user-defined trigger conditions, enabling complex coordinated behaviors such as “home mode” or “sleep mode”. However, end users often face challenges when creating customized rules, particularly when coordinating multiple devices and addressing complex intentions. On the one hand, users typically lack the technical expertise to clearly articulate their needs or translate them into formal rules. On the other hand, existing methods for generating automation rules have two primary limitations. First, they lack the capability to model complex scenarios, making it difficult to support advanced automation needs involving multi-device coordination. Second, they overly rely on users providing explicit, structured instructions, overlooking the implicit expression of user intentions. To address these issues, this study proposes HomeMind, an intelligent agent system powered by large language models (LLMs) that integrates pattern mining and semantic understanding capabilities. The workflow of HomeMind comprises three core steps. First, an unsupervised method for mining frequent event patterns is used to adaptively process event sequences of varying lengths, enabling more accurate identification of event patterns. Second, temporal features at both weekly and daily scales are extracted for each pattern, significantly enhancing its temporal context. Finally, a context-augmented prompt mechanism is leveraged to guide the LLM in inferring user intentions through chain-of-thought reasoning and generating semantically clear scenario automation rules. Ultimately, HomeMind presents candidate rules to users, facilitating the understanding and selection of automation solutions best suited to their lifestyles, and supports integration with smart home platforms for one-click rule deployment. Comprehensive evaluations of HomeMind are conducted on both real-world and synthetic datasets. Experimental results demonstrate that HomeMind outperforms existing baseline methods in both intention inference and rule generation accuracy.
In smart home systems, scenario automation rules can automatically execute a set of device operations based on user-defined trigger conditions, enabling complex coordinated behaviors such as “home mode” or “sleep mode”. However, end users often face challenges when creating customized rules, particularly when coordinating multiple devices and addressing complex intentions. On the one hand, users typically lack the technical expertise to clearly articulate their needs or translate them into formal rules. On the other hand, existing methods for generating automation rules have two primary limitations. First, they lack the capability to model complex scenarios, making it difficult to support advanced automation needs involving multi-device coordination. Second, they overly rely on users providing explicit, structured instructions, overlooking the implicit expression of user intentions. To address these issues, this study proposes HomeMind, an intelligent agent system powered by large language models (LLMs) that integrates pattern mining and semantic understanding capabilities. The workflow of HomeMind comprises three core steps. First, an unsupervised method for mining frequent event patterns is used to adaptively process event sequences of varying lengths, enabling more accurate identification of event patterns. Second, temporal features at both weekly and daily scales are extracted for each pattern, significantly enhancing its temporal context. Finally, a context-augmented prompt mechanism is leveraged to guide the LLM in inferring user intentions through chain-of-thought reasoning and generating semantically clear scenario automation rules. Ultimately, HomeMind presents candidate rules to users, facilitating the understanding and selection of automation solutions best suited to their lifestyles, and supports integration with smart home platforms for one-click rule deployment. Comprehensive evaluations of HomeMind are conducted on both real-world and synthetic datasets. Experimental results demonstrate that HomeMind outperforms existing baseline methods in both intention inference and rule generation accuracy.
Available online: June 03, 2026
, DOI: 10.13328/j.cnki.jos.007637
Abstract:
In the era of big data, traditional indexes (e.g., B+ trees) face challenges, including high memory consumption. Learned indexes are increasingly replacing traditional indexes because of their lower memory consumption and higher query efficiency. However, existing learned indexes struggle to effectively adapt to diverse data distributions and are prone to performance degradation when distribution shifts occur due to new data insertions. To address these issues, this study proposes HCA-Index, a learned index based on data hot-cold awareness. The core components of HCA-Index include 1) designing a progressive partitioning algorithm based on error thresholds to dynamically fit data distributions; 2) constructing a high-precision index by extracting key ranges in a bottom-up manner and adopting hierarchical merging strategies; 3) designing a dynamically evolving data temperature calculation model to identify hot and cold data; and 4) leveraging node-level hot-cold partitioning to support data migration and fast querying. Experiments on real-world datasets demonstrate that, compared with traditional indexes and state-of-the-art learned indexes, HCA-Index significantly reduces query latency, further lowers memory consumption, and effectively reduces the number of retraining operations caused by new data insertions.
In the era of big data, traditional indexes (e.g., B+ trees) face challenges, including high memory consumption. Learned indexes are increasingly replacing traditional indexes because of their lower memory consumption and higher query efficiency. However, existing learned indexes struggle to effectively adapt to diverse data distributions and are prone to performance degradation when distribution shifts occur due to new data insertions. To address these issues, this study proposes HCA-Index, a learned index based on data hot-cold awareness. The core components of HCA-Index include 1) designing a progressive partitioning algorithm based on error thresholds to dynamically fit data distributions; 2) constructing a high-precision index by extracting key ranges in a bottom-up manner and adopting hierarchical merging strategies; 3) designing a dynamically evolving data temperature calculation model to identify hot and cold data; and 4) leveraging node-level hot-cold partitioning to support data migration and fast querying. Experiments on real-world datasets demonstrate that, compared with traditional indexes and state-of-the-art learned indexes, HCA-Index significantly reduces query latency, further lowers memory consumption, and effectively reduces the number of retraining operations caused by new data insertions.
Available online: June 03, 2026
, DOI: 10.13328/j.cnki.jos.007654
Abstract:
Dynamic graph representation learning captures time-evolving topological structures and interaction patterns among entities, providing spatiotemporally aware embedding for downstream tasks such as link prediction, revealing the dynamic evolution patterns of complex systems. Continuous-time dynamic graphs, owing to their rich fine-grained temporal information, provide a more realistic modeling paradigm for complex processes such as social network evolution. However, current continuous-time dynamic graph representation learning faces at least three challenges. (1) The effective extraction of critical information from long historical interaction sequences requires robust modeling of long-term temporal dependencies. (2) Handling long interaction sequences requires controlling computational complexity to improve efficiency. (3) Traditional approaches that capture interaction patterns in chronological order struggle to reveal hidden non-causal associations. To address these challenges, this study proposes a novel continuous-time dynamic graph representation learning model. By leveraging a bidirectional selective state-space encoding mechanism, long-term temporal dependencies in interaction sequences can be captured, and subsequent interaction information can also be incorporated through backward pathways. The temporal unidirectional constraint is thus broken, enhancing the model’s global context comprehension capability. Extensive experimental results demonstrate that the model consistently outperforms baseline methods in predictive performance on real-world datasets spanning diverse domains, while maintaining high computational efficiency, enabling the modeling of long-term bidirectional dependencies under limited computational resources.
Dynamic graph representation learning captures time-evolving topological structures and interaction patterns among entities, providing spatiotemporally aware embedding for downstream tasks such as link prediction, revealing the dynamic evolution patterns of complex systems. Continuous-time dynamic graphs, owing to their rich fine-grained temporal information, provide a more realistic modeling paradigm for complex processes such as social network evolution. However, current continuous-time dynamic graph representation learning faces at least three challenges. (1) The effective extraction of critical information from long historical interaction sequences requires robust modeling of long-term temporal dependencies. (2) Handling long interaction sequences requires controlling computational complexity to improve efficiency. (3) Traditional approaches that capture interaction patterns in chronological order struggle to reveal hidden non-causal associations. To address these challenges, this study proposes a novel continuous-time dynamic graph representation learning model. By leveraging a bidirectional selective state-space encoding mechanism, long-term temporal dependencies in interaction sequences can be captured, and subsequent interaction information can also be incorporated through backward pathways. The temporal unidirectional constraint is thus broken, enhancing the model’s global context comprehension capability. Extensive experimental results demonstrate that the model consistently outperforms baseline methods in predictive performance on real-world datasets spanning diverse domains, while maintaining high computational efficiency, enabling the modeling of long-term bidirectional dependencies under limited computational resources.
Available online: June 01, 2026
, DOI: 10.13328/j.cnki.jos.007638
Abstract:
With the rapid development of big data and artificial intelligence technologies, graph databases have gradually become core infrastructure for social network analysis, financial risk control, and knowledge graphs due to their advantages in complex relationship modeling and efficient querying. Graph databases manage two primary objects: nodes and the relationships between them. At the architectural level, the node identifier (NodeID) serves as the critical link for graph data management, undertaking key functions including node identity representation, relationship lookup, and graph algorithm execution. Current mainstream graph databases widely adopt key-value store (KVS) to implement NodeID-to-graph-structure mapping management. However, existing systems largely rely on general-purpose KVSs (e.g., RocksDB) for managing such mappings, yet lack in-depth consideration of workload characteristics. These limitations are reflected in two aspects: (1) the lack of workload modeling for node identifier mapping, and (2) insufficient adaptability to heterogeneous software and hardware environments (e.g., CPU/memory, SSD/HDD). This study first systematically analyzes the operational characteristics and KVS requirements of node identifier mapping in graph databases. It then evaluates the performance of multiple mainstream KVS engines (including RocksDB, LMDB, LevelDB, FasterKV, and ForestDB) in heterogeneous hardware environments, systematically revealing the impact patterns of data workloads (e.g., data scale, read-write ratio) and hardware configurations (e.g., memory capacity, thread count, storage medium) on execution efficiency. Based on large-scale experiments involving five datasets, three hardware platforms, and over 1 300 comparative tests, this study proposes an adaptation strategy based on decision tree model that integrates workload characteristics (data scale and read-write ratio) with hardware configurations (e.g., memory, thread count, disk type) to guide adaptive selection of KVS engines. Experiments show that the model achieves 92.1% accuracy in recommending optimal engines, with suboptimal scenarios exhibiting less than 10% performance gap.
With the rapid development of big data and artificial intelligence technologies, graph databases have gradually become core infrastructure for social network analysis, financial risk control, and knowledge graphs due to their advantages in complex relationship modeling and efficient querying. Graph databases manage two primary objects: nodes and the relationships between them. At the architectural level, the node identifier (NodeID) serves as the critical link for graph data management, undertaking key functions including node identity representation, relationship lookup, and graph algorithm execution. Current mainstream graph databases widely adopt key-value store (KVS) to implement NodeID-to-graph-structure mapping management. However, existing systems largely rely on general-purpose KVSs (e.g., RocksDB) for managing such mappings, yet lack in-depth consideration of workload characteristics. These limitations are reflected in two aspects: (1) the lack of workload modeling for node identifier mapping, and (2) insufficient adaptability to heterogeneous software and hardware environments (e.g., CPU/memory, SSD/HDD). This study first systematically analyzes the operational characteristics and KVS requirements of node identifier mapping in graph databases. It then evaluates the performance of multiple mainstream KVS engines (including RocksDB, LMDB, LevelDB, FasterKV, and ForestDB) in heterogeneous hardware environments, systematically revealing the impact patterns of data workloads (e.g., data scale, read-write ratio) and hardware configurations (e.g., memory capacity, thread count, storage medium) on execution efficiency. Based on large-scale experiments involving five datasets, three hardware platforms, and over 1 300 comparative tests, this study proposes an adaptation strategy based on decision tree model that integrates workload characteristics (data scale and read-write ratio) with hardware configurations (e.g., memory, thread count, disk type) to guide adaptive selection of KVS engines. Experiments show that the model achieves 92.1% accuracy in recommending optimal engines, with suboptimal scenarios exhibiting less than 10% performance gap.
Available online: May 27, 2026
, DOI: 10.13328/j.cnki.jos.007664
Abstract:
The key optimization technologies of traditional relational databases generally face bottlenecks including insufficient estimation accuracy, low efficiency of optimization decisions, and poor environmental adaptability when dealing with massive data processing, complex queries, and dynamic workload scenarios. Multi-granularity granular-ball computing provides a new solution for improving the performance of database systems, showing great research potential and application prospects. The core directions of artificial intelligence in driving intelligent optimization for databases are outlined, and the main challenges faced by existing learning-based optimization methods in terms of model generalization ability, interpretability, and handling complex queries and dynamic data distributions are explored. On this basis, the current status and key technologies of database optimization are systematically reviewed. Combined with multi-granularity granular-ball computing, database optimization techniques primarily focus on two aspects: query optimization and configuration optimization. For query optimization, the key techniques include cardinality estimation and join order selection. In terms of cardinality estimation, traditional methods are difficult to effectively support accurate estimation of complex joins and nested queries involving multiple tables, and often result in huge storage overhead. Learning-based methods can better handle high-dimensional data relationships, and a new method that uses multi-granularity granular-ball computing techniques to extract hierarchical distribution features of data and combines them with tree-structured neural networks is introduced, which can significantly improve the robustness and accuracy of complex query cardinality estimation. In terms of join order selection, traditional methods have low search efficiency in multi-table join scenarios. Static methods based on historical experience learning have limited adaptability to new query patterns. Although dynamic learning methods can support runtime adjustments, they incur a high cost. In contrast, representing the join plan as a combination of multi-granularity granular-balls with geometric relationships, optimizing the search space through a hierarchical structure, and integrating deep reinforcement learning for decision-making provides an efficient approach to finding a globally approximate optimal join order. For database configuration optimization, parameter tuning plays a key role in improving performance. Traditional optimization techniques based on search are difficult to obtain the global optimal solution within a reasonable time. Although traditional machine learning methods can achieve automated tuning, they highly rely on the quality and coverage of training data. Reinforcement learning methods continuously improve strategies by interacting with the system environment, requiring only a small number of samples to achieve strong adaptive parameter tuning. By integrating multi-granularity granular-ball computing methods, the characteristics of the parameter space can be efficiently represented, thus significantly improving tuning efficiency and effectiveness. Although the prospects of optimizing intelligent databases using granular-ball computing techniques are broad, practical applications still face major challenges such as effective integration with existing database models, reduction of computational overhead, and ensuring model stability under dynamic workload changes. Future research requires continuous development of theories and technologies to promote database systems toward more intelligent, efficient, and robust directions.
The key optimization technologies of traditional relational databases generally face bottlenecks including insufficient estimation accuracy, low efficiency of optimization decisions, and poor environmental adaptability when dealing with massive data processing, complex queries, and dynamic workload scenarios. Multi-granularity granular-ball computing provides a new solution for improving the performance of database systems, showing great research potential and application prospects. The core directions of artificial intelligence in driving intelligent optimization for databases are outlined, and the main challenges faced by existing learning-based optimization methods in terms of model generalization ability, interpretability, and handling complex queries and dynamic data distributions are explored. On this basis, the current status and key technologies of database optimization are systematically reviewed. Combined with multi-granularity granular-ball computing, database optimization techniques primarily focus on two aspects: query optimization and configuration optimization. For query optimization, the key techniques include cardinality estimation and join order selection. In terms of cardinality estimation, traditional methods are difficult to effectively support accurate estimation of complex joins and nested queries involving multiple tables, and often result in huge storage overhead. Learning-based methods can better handle high-dimensional data relationships, and a new method that uses multi-granularity granular-ball computing techniques to extract hierarchical distribution features of data and combines them with tree-structured neural networks is introduced, which can significantly improve the robustness and accuracy of complex query cardinality estimation. In terms of join order selection, traditional methods have low search efficiency in multi-table join scenarios. Static methods based on historical experience learning have limited adaptability to new query patterns. Although dynamic learning methods can support runtime adjustments, they incur a high cost. In contrast, representing the join plan as a combination of multi-granularity granular-balls with geometric relationships, optimizing the search space through a hierarchical structure, and integrating deep reinforcement learning for decision-making provides an efficient approach to finding a globally approximate optimal join order. For database configuration optimization, parameter tuning plays a key role in improving performance. Traditional optimization techniques based on search are difficult to obtain the global optimal solution within a reasonable time. Although traditional machine learning methods can achieve automated tuning, they highly rely on the quality and coverage of training data. Reinforcement learning methods continuously improve strategies by interacting with the system environment, requiring only a small number of samples to achieve strong adaptive parameter tuning. By integrating multi-granularity granular-ball computing methods, the characteristics of the parameter space can be efficiently represented, thus significantly improving tuning efficiency and effectiveness. Although the prospects of optimizing intelligent databases using granular-ball computing techniques are broad, practical applications still face major challenges such as effective integration with existing database models, reduction of computational overhead, and ensuring model stability under dynamic workload changes. Future research requires continuous development of theories and technologies to promote database systems toward more intelligent, efficient, and robust directions.
Available online: May 27, 2026
, DOI: 10.13328/j.cnki.jos.007665
Abstract:
Federated learning (FL), as a distributed machine learning method, enables model training while protecting user privacy and data security. However, the involvement of multiple parties and the widespread exposure of models in FL can easily lead to copyright leakage. This study proposes a watermarking scheme with ownership verification, model leakage tracing, and lazy client detection. The proposed scheme introduces a client identity-based backdoor watermark generation mechanism and federated dynamic weight adjustment (FDWA) to ensure the uniqueness of each client’s watermark and resolve watermark conflicts. Model fidelity and watermark trigger rates are significantly improved, while also achieving better performance in detecting lazy clients. Experimental results show that the scheme provides more comprehensive protection while maintaining model performance, significantly improves watermark trigger rates, and effectively resists various attacks such as fine-tuning, pruning, quantization, and collusion attacks, thus enhancing the security and fairness of the FL environment and providing effective copyright protection for models.
Federated learning (FL), as a distributed machine learning method, enables model training while protecting user privacy and data security. However, the involvement of multiple parties and the widespread exposure of models in FL can easily lead to copyright leakage. This study proposes a watermarking scheme with ownership verification, model leakage tracing, and lazy client detection. The proposed scheme introduces a client identity-based backdoor watermark generation mechanism and federated dynamic weight adjustment (FDWA) to ensure the uniqueness of each client’s watermark and resolve watermark conflicts. Model fidelity and watermark trigger rates are significantly improved, while also achieving better performance in detecting lazy clients. Experimental results show that the scheme provides more comprehensive protection while maintaining model performance, significantly improves watermark trigger rates, and effectively resists various attacks such as fine-tuning, pruning, quantization, and collusion attacks, thus enhancing the security and fairness of the FL environment and providing effective copyright protection for models.
Available online: May 27, 2026
, DOI: 10.13328/j.cnki.jos.007640
Abstract:
Commit messages are natural language text that describe code changes and are crucial for understanding code and its evolution. Constrained by software development costs and related factors, developers often fail to carefully craft commit messages, resulting in unsatisfactory quality in existing software projects. As a result, commit message generation has gained widespread attention. Current work mainly involves collecting commit message data from open-source software projects and training deep learning models on this data to accomplish the task. However, due to the quality issues of open-source project data, existing methods struggle to achieve satisfactory performance. Large language models, by pre-training on vast amounts of code and text data, learn rich linguistic patterns and world knowledge, enabling the generation of high-quality, natural, and contextually appropriate text, thus providing new directions for commit message generation. This study proposes a commit message generation method based on large language models, in which large language models are applied to commit message generation through techniques such as in-context learning and model fine-tuning. Two example retrieval methods are employed to enhance the in-context learning approach. The generated text is then analyzed in terms of naturalness and relevance, and the strengths and limitations of the large language model-based methods, as well as ways to address these limitations, are examined. Experiments demonstrate that large language models outperform baseline methods. In addition, chain-of-thought reasoning and larger-scale models can address the issue of insufficient understanding of code changes.
Commit messages are natural language text that describe code changes and are crucial for understanding code and its evolution. Constrained by software development costs and related factors, developers often fail to carefully craft commit messages, resulting in unsatisfactory quality in existing software projects. As a result, commit message generation has gained widespread attention. Current work mainly involves collecting commit message data from open-source software projects and training deep learning models on this data to accomplish the task. However, due to the quality issues of open-source project data, existing methods struggle to achieve satisfactory performance. Large language models, by pre-training on vast amounts of code and text data, learn rich linguistic patterns and world knowledge, enabling the generation of high-quality, natural, and contextually appropriate text, thus providing new directions for commit message generation. This study proposes a commit message generation method based on large language models, in which large language models are applied to commit message generation through techniques such as in-context learning and model fine-tuning. Two example retrieval methods are employed to enhance the in-context learning approach. The generated text is then analyzed in terms of naturalness and relevance, and the strengths and limitations of the large language model-based methods, as well as ways to address these limitations, are examined. Experiments demonstrate that large language models outperform baseline methods. In addition, chain-of-thought reasoning and larger-scale models can address the issue of insufficient understanding of code changes.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007639
Abstract:
The adaptability of deep neural network (DNN) models in non-stationary scenarios remains a significant challenge in the field of artificial intelligence. In particular, when concept drift occurs, models with predefined architectural parameters struggle to adapt to evolving data distributions. Existing depth adjustment methods for DNNs do not adequately evaluate the validity of depth expansion and overlook the synergy between network weights and depth during the adjustment process. To address this issue, this study proposes an adaptive online deep neural network (AODNN). AODNN assesses the effectiveness of depth expansion by jointly analyzing the network’s loss trends, classifier weights, and changes in mutual information, thus enabling adaptive depth growth. In addition, it employs parameter update optimization to select intermediate classifiers for participation in backpropagation, reducing interference from deeper intermediate classifiers to shallower ones and accelerating convergence. Comparative experiments are conducted between AODNN and state-of-the-art methods on multiple real-world and synthetic datasets containing concept drift. The experimental results validate the effectiveness of AODNN’s adaptive depth growth and parameter update optimization strategies, demonstrating its capability to effectively capture changes in data distribution and mitigate the impact of concept drift. Across key performance metrics, AODNN demonstrates significant advantages, notably outperforming HBP, ANSN, and EODL in cumulative accuracy, surpassing ATNN, ANSN, and EODL in F1-score, and achieving significantly better MCC compared to ANSN and EODL. Furthermore, AODNN surpasses current state-of-the-art methods in aspects of model convergence speed.
The adaptability of deep neural network (DNN) models in non-stationary scenarios remains a significant challenge in the field of artificial intelligence. In particular, when concept drift occurs, models with predefined architectural parameters struggle to adapt to evolving data distributions. Existing depth adjustment methods for DNNs do not adequately evaluate the validity of depth expansion and overlook the synergy between network weights and depth during the adjustment process. To address this issue, this study proposes an adaptive online deep neural network (AODNN). AODNN assesses the effectiveness of depth expansion by jointly analyzing the network’s loss trends, classifier weights, and changes in mutual information, thus enabling adaptive depth growth. In addition, it employs parameter update optimization to select intermediate classifiers for participation in backpropagation, reducing interference from deeper intermediate classifiers to shallower ones and accelerating convergence. Comparative experiments are conducted between AODNN and state-of-the-art methods on multiple real-world and synthetic datasets containing concept drift. The experimental results validate the effectiveness of AODNN’s adaptive depth growth and parameter update optimization strategies, demonstrating its capability to effectively capture changes in data distribution and mitigate the impact of concept drift. Across key performance metrics, AODNN demonstrates significant advantages, notably outperforming HBP, ANSN, and EODL in cumulative accuracy, surpassing ATNN, ANSN, and EODL in F1-score, and achieving significantly better MCC compared to ANSN and EODL. Furthermore, AODNN surpasses current state-of-the-art methods in aspects of model convergence speed.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007660
Abstract:
As a privacy-preserving authentication technique, ring signatures preserve user anonymity while ensuring message authenticity and integrity. They are widely used in scenarios such as electronic voting and blockchain transactions. To prevent malicious users from abusing anonymity and engaging in dishonest behavior, existing ring signature schemes employ a tracing authority (or regulatory authority) to reveal the real identities of any user. However, introducing such an authority may lead to privacy leakage for honest users, disrupting the balance between anonymity and traceability. To address these issues and promote the application and development of domestic cryptographic technologies, this study proposes an SM2-based ring signature scheme with traceability for illegal messages. The proposed scheme designs a tracing mechanism for a single illegal message, under which the tracing authority is permitted to reveal only the real identities of malicious users who publish illegal messages, while honest users remain anonymous, thus balancing anonymity and traceability. On this basis, this study further extends the proposed scheme by combining polynomial techniques with Merkle hash tree techniques. As a result, traceability for an arbitrary number of illegal messages is achieved, the scope of supervision over malicious users is expanded, and user verification overhead and communication costs are reduced. Security analysis demonstrates that the proposed scheme achieves unforgeability, anonymity, and traceability. Performance analysis indicates that the proposed scheme offers superior efficiency and practicality compared with existing ring signature schemes.
As a privacy-preserving authentication technique, ring signatures preserve user anonymity while ensuring message authenticity and integrity. They are widely used in scenarios such as electronic voting and blockchain transactions. To prevent malicious users from abusing anonymity and engaging in dishonest behavior, existing ring signature schemes employ a tracing authority (or regulatory authority) to reveal the real identities of any user. However, introducing such an authority may lead to privacy leakage for honest users, disrupting the balance between anonymity and traceability. To address these issues and promote the application and development of domestic cryptographic technologies, this study proposes an SM2-based ring signature scheme with traceability for illegal messages. The proposed scheme designs a tracing mechanism for a single illegal message, under which the tracing authority is permitted to reveal only the real identities of malicious users who publish illegal messages, while honest users remain anonymous, thus balancing anonymity and traceability. On this basis, this study further extends the proposed scheme by combining polynomial techniques with Merkle hash tree techniques. As a result, traceability for an arbitrary number of illegal messages is achieved, the scope of supervision over malicious users is expanded, and user verification overhead and communication costs are reduced. Security analysis demonstrates that the proposed scheme achieves unforgeability, anonymity, and traceability. Performance analysis indicates that the proposed scheme offers superior efficiency and practicality compared with existing ring signature schemes.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007636
Abstract:
Federated learning is currently the most advanced technology in secure multi-party machine learning and has attracted significant attention for its ability to enable collaborative model training across multiple nodes without requiring data to leave local devices. However, in real-world scenarios, client data are often non-independent and identically distributed (Non-IID), which significantly degrades the performance of the global model. Current mainstream methods primarily focus on correcting model parameter biases during training, but their effectiveness diminishes as data distribution becomes more extreme. Another category of methods aims to correct data distribution biases before local training, but this approach poses privacy risks. This study proposes a framework that adheres to strict federated learning privacy protection requirements and utilizes techniques such as differential privacy and secure multi-party computation. Assuming that client data distributions remain black boxes, the most beneficial labels for improving global model performance are selected. High-quality synthetic data are generated for these labels and distributed to clients, thus making the local data distribution closer to IID and significantly enhancing the performance of the global model. Specifically, a privacy-preserving algorithm is first designed to capture the distribution of labels across clients, and clients are grouped according to these label distribution characteristics. Then, the most worthwhile labels for generation are selected, and within the corresponding groups, clients are coordinated to collaboratively train high-quality global generative models, which are stored on the server. During the training phase of the task model, based on label distribution characteristics, samples are selectively synthesized using these generative models and distributed to clients, resulting in a more uniform local data distribution. This process reduces local model disparities and yields a high-quality global model after aggregation. Experimental results demonstrate that the proposed method can effectively improve the accuracy of the global model on the test set and reduce the number of federated learning communication rounds required before convergence. Its effectiveness surpasses that of baseline methods and has been validated on different datasets.
Federated learning is currently the most advanced technology in secure multi-party machine learning and has attracted significant attention for its ability to enable collaborative model training across multiple nodes without requiring data to leave local devices. However, in real-world scenarios, client data are often non-independent and identically distributed (Non-IID), which significantly degrades the performance of the global model. Current mainstream methods primarily focus on correcting model parameter biases during training, but their effectiveness diminishes as data distribution becomes more extreme. Another category of methods aims to correct data distribution biases before local training, but this approach poses privacy risks. This study proposes a framework that adheres to strict federated learning privacy protection requirements and utilizes techniques such as differential privacy and secure multi-party computation. Assuming that client data distributions remain black boxes, the most beneficial labels for improving global model performance are selected. High-quality synthetic data are generated for these labels and distributed to clients, thus making the local data distribution closer to IID and significantly enhancing the performance of the global model. Specifically, a privacy-preserving algorithm is first designed to capture the distribution of labels across clients, and clients are grouped according to these label distribution characteristics. Then, the most worthwhile labels for generation are selected, and within the corresponding groups, clients are coordinated to collaboratively train high-quality global generative models, which are stored on the server. During the training phase of the task model, based on label distribution characteristics, samples are selectively synthesized using these generative models and distributed to clients, resulting in a more uniform local data distribution. This process reduces local model disparities and yields a high-quality global model after aggregation. Experimental results demonstrate that the proposed method can effectively improve the accuracy of the global model on the test set and reduce the number of federated learning communication rounds required before convergence. Its effectiveness surpasses that of baseline methods and has been validated on different datasets.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007633
Abstract:
Learning the causal structure among latent variables is a key technique for revealing the underlying mechanisms of phenomena in scientific research, with its core objective being to infer causal relationships between latent variables from observational data. Existing methods generally rely on the “pure child” assumption, which assumes that no direct causal connections exist among the observed child variables (measurement variables) corresponding to latent variables. However, this assumption often fails to hold in many real-world scenarios, thus limiting the identifiability of existing methods. To address this challenge, this study investigates the identifiability of latent variables under the “non-pure” measurement setting and proposes a linear non-Gaussian acyclic latent variable model (LiNGLM) that allows causal relationships among observed variables. Based on this model, a latent variable structure learning (LLSTIN) algorithm is proposed. This algorithm is based on the transformed independent noise (TIN) condition and the graph criterion established thereby. First, the “single-factor sets” corresponding to each latent variable are extracted through the build causal cluster (BCC) algorithm to identify the existence of latent variables. Then, the “root” observed variables in the sets are selected as effective proxies for latent variables, and the causal relationships between latent variables are further identified. This study theoretically proves that the proposed algorithm can correctly extract single-factor sets and further identify the causal structure among latent variables. Simulation data and real-world data experimental results further verify the correctness and effectiveness of the proposed algorithm.
Learning the causal structure among latent variables is a key technique for revealing the underlying mechanisms of phenomena in scientific research, with its core objective being to infer causal relationships between latent variables from observational data. Existing methods generally rely on the “pure child” assumption, which assumes that no direct causal connections exist among the observed child variables (measurement variables) corresponding to latent variables. However, this assumption often fails to hold in many real-world scenarios, thus limiting the identifiability of existing methods. To address this challenge, this study investigates the identifiability of latent variables under the “non-pure” measurement setting and proposes a linear non-Gaussian acyclic latent variable model (LiNGLM) that allows causal relationships among observed variables. Based on this model, a latent variable structure learning (LLSTIN) algorithm is proposed. This algorithm is based on the transformed independent noise (TIN) condition and the graph criterion established thereby. First, the “single-factor sets” corresponding to each latent variable are extracted through the build causal cluster (BCC) algorithm to identify the existence of latent variables. Then, the “root” observed variables in the sets are selected as effective proxies for latent variables, and the causal relationships between latent variables are further identified. This study theoretically proves that the proposed algorithm can correctly extract single-factor sets and further identify the causal structure among latent variables. Simulation data and real-world data experimental results further verify the correctness and effectiveness of the proposed algorithm.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007634
Abstract:
In recent years, smart contracts have been widely used in the financial field due to their immutability and enforceability. At the same time, the number of security incidents caused by smart contracts has continued to increase, often causing large-scale economic losses. Therefore, most researchers focus on developing vulnerability detection tools to assess the security of smart contracts. However, the performance of different vulnerability detection tools cannot be objectively evaluated due to the use of inconsistent datasets. This study constructs a new dataset and systematically tests nine candidate tools under a unified benchmark. The dataset includes real-world smart contracts and covers five common types of vulnerabilities. This study evaluates the tools from five aspects and proposes a new smart contract classification method to verify the robustness of the tools. The experimental results are as follows. 1) Most existing tools are easy to install, but there are also problems such as discontinued maintenance. 2) Static detection tools still face the problem of high false positive and false negative rates in the actual detection process, and rely on the analysis of different versions of smart contracts. 3) Static detection tools have a small time overhead, while tools based on symbolic execution are prone to large time overhead due to state explosion. 4) Some static detection tools do not support contracts with complex inheritance relationships. 5) The integration of multiple vulnerability detection technologies can effectively improve the detection precision and recall.
In recent years, smart contracts have been widely used in the financial field due to their immutability and enforceability. At the same time, the number of security incidents caused by smart contracts has continued to increase, often causing large-scale economic losses. Therefore, most researchers focus on developing vulnerability detection tools to assess the security of smart contracts. However, the performance of different vulnerability detection tools cannot be objectively evaluated due to the use of inconsistent datasets. This study constructs a new dataset and systematically tests nine candidate tools under a unified benchmark. The dataset includes real-world smart contracts and covers five common types of vulnerabilities. This study evaluates the tools from five aspects and proposes a new smart contract classification method to verify the robustness of the tools. The experimental results are as follows. 1) Most existing tools are easy to install, but there are also problems such as discontinued maintenance. 2) Static detection tools still face the problem of high false positive and false negative rates in the actual detection process, and rely on the analysis of different versions of smart contracts. 3) Static detection tools have a small time overhead, while tools based on symbolic execution are prone to large time overhead due to state explosion. 4) Some static detection tools do not support contracts with complex inheritance relationships. 5) The integration of multiple vulnerability detection technologies can effectively improve the detection precision and recall.
Available online: May 20, 2026
, DOI: 10.13328/j.cnki.jos.007623
Abstract:
Object storage is one of the most widely used storage solutions in cloud computing environments. Replicating data stored in object storage across one or more cloud regions improves reliability and availability while reducing access latency for users. An object storage replication system is required to ensure consistency among replicas while minimizing replication latency. Existing open-source cross-cloud object storage replication systems, such as Skyplane, support object data replication across major cloud platforms but do not guarantee data consistency and exhibit high replication latency under dynamic workloads. To enable low-cost and low-latency object data replication under dynamic workloads, this study proposes a cross-cloud object storage replication system, termed FastRep, based on serverless computing. Data consistency is ensured through object-grained replication task locks and an optimistic replication mechanism. Cloud function invocation latency is reduced via a time-bounded complete tree, while bandwidth degradation is mitigated through adaptive function cloning. Experimental results demonstrate that FastRep achieves low-cost replication of dynamically sized object data with latency at the minute level.
Object storage is one of the most widely used storage solutions in cloud computing environments. Replicating data stored in object storage across one or more cloud regions improves reliability and availability while reducing access latency for users. An object storage replication system is required to ensure consistency among replicas while minimizing replication latency. Existing open-source cross-cloud object storage replication systems, such as Skyplane, support object data replication across major cloud platforms but do not guarantee data consistency and exhibit high replication latency under dynamic workloads. To enable low-cost and low-latency object data replication under dynamic workloads, this study proposes a cross-cloud object storage replication system, termed FastRep, based on serverless computing. Data consistency is ensured through object-grained replication task locks and an optimistic replication mechanism. Cloud function invocation latency is reduced via a time-bounded complete tree, while bandwidth degradation is mitigated through adaptive function cloning. Experimental results demonstrate that FastRep achieves low-cost replication of dynamically sized object data with latency at the minute level.
Available online: May 13, 2026
, DOI: 10.13328/j.cnki.jos.007635
Abstract:
Existing fuzzer for browser JavaScript engines still have limitations in detecting potential vulnerabilities, especially in identifying non-crashing heap memory errors. This study proposes ToxiHeap, a JavaScript engine fuzzing framework that integrates runtime monitoring based on memory toxic labeling with large language model (LLM)-guided semantic generation as two parallel core components. The detection track instruments heap allocation and access paths in the target engine. Shadow memory and fine-grained toxic labeling provide a unified mechanism for representing multiple heap errors, including use-after-free (UAF), double free, heap out-of-bounds reads and writes (OOB-R/W), memory leak, and uninitialized memory read (UMR). Accesses to freed memory and other illegal states are converted into consumable exception signals, which compensates for the lack of effective detection signals for non-crashing vulnerabilities in approaches that rely solely on crash signals. The generation track employs a two-stage “distill-reward” LLM-based mutator that extracts semantic features from historical proofs of concept (PoCs) and generates state-dependent test cases in a targeted manner, thus overcoming the limitations of purely syntax-based or shallow-semantic mutations in exploring deep execution paths. The two tracks are integrated through a feedback loop driven by coverage gains and non-crashing detection signals. This feedback loop jointly drives sample retention, path weight updates, and knowledge base growth. An isolated replay verification mechanism is also introduced to revalidate anomalies in a clean process, which significantly reduces nondeterministic false positives. Experimental results show that on the JavaScriptCore (JSC) engine, ToxiHeap increases branch coverage from 31.17% to 33.52% in a 24-hour run, and achieves the highest or near-highest branch coverage on other mainstream engines including V8, SpiderMonkey (SM), and ChakraCore (CH). The proportion of valid samples remains above 91%. On a set of 50 PoC references covering multiple vulnerability patterns including UAF, the average detection rate across the four engines reaches 89.18%.
Existing fuzzer for browser JavaScript engines still have limitations in detecting potential vulnerabilities, especially in identifying non-crashing heap memory errors. This study proposes ToxiHeap, a JavaScript engine fuzzing framework that integrates runtime monitoring based on memory toxic labeling with large language model (LLM)-guided semantic generation as two parallel core components. The detection track instruments heap allocation and access paths in the target engine. Shadow memory and fine-grained toxic labeling provide a unified mechanism for representing multiple heap errors, including use-after-free (UAF), double free, heap out-of-bounds reads and writes (OOB-R/W), memory leak, and uninitialized memory read (UMR). Accesses to freed memory and other illegal states are converted into consumable exception signals, which compensates for the lack of effective detection signals for non-crashing vulnerabilities in approaches that rely solely on crash signals. The generation track employs a two-stage “distill-reward” LLM-based mutator that extracts semantic features from historical proofs of concept (PoCs) and generates state-dependent test cases in a targeted manner, thus overcoming the limitations of purely syntax-based or shallow-semantic mutations in exploring deep execution paths. The two tracks are integrated through a feedback loop driven by coverage gains and non-crashing detection signals. This feedback loop jointly drives sample retention, path weight updates, and knowledge base growth. An isolated replay verification mechanism is also introduced to revalidate anomalies in a clean process, which significantly reduces nondeterministic false positives. Experimental results show that on the JavaScriptCore (JSC) engine, ToxiHeap increases branch coverage from 31.17% to 33.52% in a 24-hour run, and achieves the highest or near-highest branch coverage on other mainstream engines including V8, SpiderMonkey (SM), and ChakraCore (CH). The proportion of valid samples remains above 91%. On a set of 50 PoC references covering multiple vulnerability patterns including UAF, the average detection rate across the four engines reaches 89.18%.
Available online: May 13, 2026
, DOI: 10.13328/j.cnki.jos.007644
Abstract:
Compiler optimization options serve as a core interface between the compiler and developers, and their proper configuration directly impacts software quality and development efficiency. However, there are a large number of options with complex interrelations, and their optimization effects are dynamically influenced by program features and target machines. The official documentation is also insufficient to provide the configuration guidance needed for practical application scenarios. To address these challenges, this study combines static and dynamic analysis techniques applied to the GCC source code and investigates two types of relationships that affect the effectiveness of compiler options: internal constraint relationships and external environment dependencies. First, 68 pairs of internal constraint relationships (involving 84 options) and 103 pairs of external environment dependencies (including 51 architecture-sensitive options and 52 program feature-sensitive options) are identified through static analysis, and the effectiveness of these options in different external environments is further validated through dynamic analysis. Finally, the analysis results of option relationships are applied to three compiler-related scenarios: 1) improving the effectiveness of compiler auto-tuning by designing a relationship-aware genetic auto-tuning algorithm, which achieves an average performance improvement of 3.65% compared with algorithms that do not consider option relationships, such as the genetic algorithm; 2) improving the effectiveness of compiler testing by proposing a option sequence test case generation method based on option relationships, achieving a 51.7% increase in generation efficiency compared to random methods; 3) providing developers with more effective option configuration schemes to help them use options properly to achieve their desired optimization goals. These three applications demonstrate the key role of compiler option relationships in compiler auto-tuning, compiler testing, and practical configuration.
Compiler optimization options serve as a core interface between the compiler and developers, and their proper configuration directly impacts software quality and development efficiency. However, there are a large number of options with complex interrelations, and their optimization effects are dynamically influenced by program features and target machines. The official documentation is also insufficient to provide the configuration guidance needed for practical application scenarios. To address these challenges, this study combines static and dynamic analysis techniques applied to the GCC source code and investigates two types of relationships that affect the effectiveness of compiler options: internal constraint relationships and external environment dependencies. First, 68 pairs of internal constraint relationships (involving 84 options) and 103 pairs of external environment dependencies (including 51 architecture-sensitive options and 52 program feature-sensitive options) are identified through static analysis, and the effectiveness of these options in different external environments is further validated through dynamic analysis. Finally, the analysis results of option relationships are applied to three compiler-related scenarios: 1) improving the effectiveness of compiler auto-tuning by designing a relationship-aware genetic auto-tuning algorithm, which achieves an average performance improvement of 3.65% compared with algorithms that do not consider option relationships, such as the genetic algorithm; 2) improving the effectiveness of compiler testing by proposing a option sequence test case generation method based on option relationships, achieving a 51.7% increase in generation efficiency compared to random methods; 3) providing developers with more effective option configuration schemes to help them use options properly to achieve their desired optimization goals. These three applications demonstrate the key role of compiler option relationships in compiler auto-tuning, compiler testing, and practical configuration.
Available online: April 29, 2026
, DOI: 10.13328/j.cnki.jos.007645
Abstract:
Edge computing has been widely adopted across various domains due to its low latency and high processing efficiency. As core foundational software in edge intelligence scenarios, the runtime reliability of the KubeEdge platform is critically important. However, software on edge systems may experience software aging after prolonged operation, leading to delayed system responses or even service interruptions, which may negatively impact user experience and potentially cause accidents. While rejuvenation operations can mitigate aging effects, current research on aging in edge systems remains relatively limited, and existing rejuvenation methods cannot be directly applied to edge systems. To address these challenges, this study proposes a comprehensive aging state determination and rejuvenation method named GIP-MI for the KubeEdge edge system. The method first employs the GCN-Informer approach to model spatial correlations and temporal dependencies among system metrics, enabling more accurate and stable predictions of future trends in these metrics compared to conventional methods. The predicted data is then fed into the deep learning method ParNet, which leverages multi-timepoint slicing and multi-resolution feature fusion to achieve more precise identification of dynamic aging states of system resources. Finally, a task offloading method based on the multi-objective evolutionary algorithm based on decomposition and information feedback model (MOEA/D-IFM) is introduced as a rejuvenation mechanism, effectively avoiding system downtime and ensuring service continuity. Experimental results demonstrate that GIP-MI outperforms baseline methods in both aging prediction and state identification accuracy. Moreover, compared to traditional rejuvenation approaches, it shows significant advantages in key metrics such as downtime, enabling effective recovery of system states
Edge computing has been widely adopted across various domains due to its low latency and high processing efficiency. As core foundational software in edge intelligence scenarios, the runtime reliability of the KubeEdge platform is critically important. However, software on edge systems may experience software aging after prolonged operation, leading to delayed system responses or even service interruptions, which may negatively impact user experience and potentially cause accidents. While rejuvenation operations can mitigate aging effects, current research on aging in edge systems remains relatively limited, and existing rejuvenation methods cannot be directly applied to edge systems. To address these challenges, this study proposes a comprehensive aging state determination and rejuvenation method named GIP-MI for the KubeEdge edge system. The method first employs the GCN-Informer approach to model spatial correlations and temporal dependencies among system metrics, enabling more accurate and stable predictions of future trends in these metrics compared to conventional methods. The predicted data is then fed into the deep learning method ParNet, which leverages multi-timepoint slicing and multi-resolution feature fusion to achieve more precise identification of dynamic aging states of system resources. Finally, a task offloading method based on the multi-objective evolutionary algorithm based on decomposition and information feedback model (MOEA/D-IFM) is introduced as a rejuvenation mechanism, effectively avoiding system downtime and ensuring service continuity. Experimental results demonstrate that GIP-MI outperforms baseline methods in both aging prediction and state identification accuracy. Moreover, compared to traditional rejuvenation approaches, it shows significant advantages in key metrics such as downtime, enabling effective recovery of system states
Available online: April 29, 2026
, DOI: 10.13328/j.cnki.jos.007641
Abstract:
Cross-chain asset swaps are a typical mode of asset circulation across blockchains. Existing cross-chain swap solutions, represented by the hashed timelock contract (HTLC) protocol, typically adopt a bind-then-lock mode, in which the follower is bound first and then locks assets. Under this mode, once the bound follower withdraws, the initiator cannot replace the swap counterparty and must wait for the timelock to expire before reclaiming the assets and restarting the swap process, which significantly prolongs the swap duration and incurs high on-chain costs. To address this issue, this study proposes a lock-then-bind mode for cross-chain asset swaps. In this paradigm, the follower is not pre-specified in the contract before the swap starts. Instead, an eligible responder first locks assets and is then bound as the follower. This paradigm avoids repeated retries and long-term asset locking caused by follower withdrawal. Under the new paradigm, this study proposes a cross-chain asset swap protocol based on an address signature lock, namely OpenSwap. OpenSwap designs an address signature lock by embedding the follower’s identity information into the lock structure, enabling it to be synchronized between two blockchains during the asset locking and unlocking process, thus ensuring consistency in follower binding across both chains. In addition, OpenSwap enhances protocol security and execution efficiency through the introduction of a challenge period and an assisted unlocking mechanism. Theoretical analysis and experimental results demonstrate that OpenSwap ensures atomicity while significantly reducing swap latency and lowering on-chain costs in low user response scenarios, providing a more flexible and efficient solution for cross-chain swaps.
Cross-chain asset swaps are a typical mode of asset circulation across blockchains. Existing cross-chain swap solutions, represented by the hashed timelock contract (HTLC) protocol, typically adopt a bind-then-lock mode, in which the follower is bound first and then locks assets. Under this mode, once the bound follower withdraws, the initiator cannot replace the swap counterparty and must wait for the timelock to expire before reclaiming the assets and restarting the swap process, which significantly prolongs the swap duration and incurs high on-chain costs. To address this issue, this study proposes a lock-then-bind mode for cross-chain asset swaps. In this paradigm, the follower is not pre-specified in the contract before the swap starts. Instead, an eligible responder first locks assets and is then bound as the follower. This paradigm avoids repeated retries and long-term asset locking caused by follower withdrawal. Under the new paradigm, this study proposes a cross-chain asset swap protocol based on an address signature lock, namely OpenSwap. OpenSwap designs an address signature lock by embedding the follower’s identity information into the lock structure, enabling it to be synchronized between two blockchains during the asset locking and unlocking process, thus ensuring consistency in follower binding across both chains. In addition, OpenSwap enhances protocol security and execution efficiency through the introduction of a challenge period and an assisted unlocking mechanism. Theoretical analysis and experimental results demonstrate that OpenSwap ensures atomicity while significantly reducing swap latency and lowering on-chain costs in low user response scenarios, providing a more flexible and efficient solution for cross-chain swaps.
Available online: April 29, 2026
, DOI: 10.13328/j.cnki.jos.007625
Abstract:
Database systems serve as critical infrastructure for big data, with their performance directly affecting the quality of service (QoS) for upper-layer applications. With the rapid advancement of storage hardware technologies, traditional database systems increasingly exhibit stability challenges under large-scale and high-concurrency workloads. Baseline evaluations reveal severe performance degradation under intensive workloads, characterized by sharp throughput decline and noticeable jitter. Analysis of key operational metrics identifies the root cause of instability in the database I/O read-write model, where delayed dirty page flushing and insufficient candidate pages for backend threads reduce overall system stability. To address these issues, this study abstracts the database I/O mechanism into a producer-consumer model centered on buffer management and identifies inherent functional coupling problems. A novel functionally decoupled I/O model is proposed, featuring optimizations to the dirty page flushing mechanism and candidate page allocation strategy, thereby enhancing the supply of clean pages. Furthermore, the improved NSGA-II algorithm is integrated into the flushing framework for multi-objective white-box parameter tuning. Comprehensive evaluations are conducted using TPC-C and sysbench benchmarks across multiple dimensions, including data scale, testing duration, concurrency levels, and read-write patterns, supplemented by ablation studies. Experimental results demonstrate that the proposed framework achieves significant improvements over baseline approaches and existing optimization approaches in throughput, stability, and latency under high-pressure scenarios.
Database systems serve as critical infrastructure for big data, with their performance directly affecting the quality of service (QoS) for upper-layer applications. With the rapid advancement of storage hardware technologies, traditional database systems increasingly exhibit stability challenges under large-scale and high-concurrency workloads. Baseline evaluations reveal severe performance degradation under intensive workloads, characterized by sharp throughput decline and noticeable jitter. Analysis of key operational metrics identifies the root cause of instability in the database I/O read-write model, where delayed dirty page flushing and insufficient candidate pages for backend threads reduce overall system stability. To address these issues, this study abstracts the database I/O mechanism into a producer-consumer model centered on buffer management and identifies inherent functional coupling problems. A novel functionally decoupled I/O model is proposed, featuring optimizations to the dirty page flushing mechanism and candidate page allocation strategy, thereby enhancing the supply of clean pages. Furthermore, the improved NSGA-II algorithm is integrated into the flushing framework for multi-objective white-box parameter tuning. Comprehensive evaluations are conducted using TPC-C and sysbench benchmarks across multiple dimensions, including data scale, testing duration, concurrency levels, and read-write patterns, supplemented by ablation studies. Experimental results demonstrate that the proposed framework achieves significant improvements over baseline approaches and existing optimization approaches in throughput, stability, and latency under high-pressure scenarios.
Available online: April 29, 2026
, DOI: 10.13328/j.cnki.jos.007622
Abstract:
Federated learning enables numerous clients to collaboratively train a model using their local data without exposing the raw data of individual parties, thereby avoiding data leakage and misuse caused by data migration in traditional machine learning paradigms. However, in practical scenarios, clients often exhibit heterogeneous data distributions and diverse system capabilities, which degrade model performance and training efficiency. Selecting a high-quality subset of clients as participants in federated learning can effectively improve global model performance and accelerate convergence. Nevertheless, existing studies show that malicious adversaries can exploit information such as local training losses or gradients to infer sensitive private data, while current participant selection strategies for heterogeneous federated learning fail to adequately address such privacy leakage risks. To tackle this challenge, this study designs a privacy-preserving participant selection protocol for heterogeneous federated learning based on secure multi-party computation. By leveraging three-party computation (3PC) secret-sharing techniques, the proposed protocol ensures data privacy during training while maintaining the accuracy of the jointly trained model. Furthermore, a secure top-k search protocol is introduced to prevent privacy leakage during the participant selection phase. The security of the proposed protocol is formally analyzed, demonstrating that the required security properties are satisfied. Experimental results indicate that, compared with heterogeneous federated learning schemes without privacy preservation, the proposed approach increases the average computational and communication overhead across all parties by 2.09%.
Federated learning enables numerous clients to collaboratively train a model using their local data without exposing the raw data of individual parties, thereby avoiding data leakage and misuse caused by data migration in traditional machine learning paradigms. However, in practical scenarios, clients often exhibit heterogeneous data distributions and diverse system capabilities, which degrade model performance and training efficiency. Selecting a high-quality subset of clients as participants in federated learning can effectively improve global model performance and accelerate convergence. Nevertheless, existing studies show that malicious adversaries can exploit information such as local training losses or gradients to infer sensitive private data, while current participant selection strategies for heterogeneous federated learning fail to adequately address such privacy leakage risks. To tackle this challenge, this study designs a privacy-preserving participant selection protocol for heterogeneous federated learning based on secure multi-party computation. By leveraging three-party computation (3PC) secret-sharing techniques, the proposed protocol ensures data privacy during training while maintaining the accuracy of the jointly trained model. Furthermore, a secure top-k search protocol is introduced to prevent privacy leakage during the participant selection phase. The security of the proposed protocol is formally analyzed, demonstrating that the required security properties are satisfied. Experimental results indicate that, compared with heterogeneous federated learning schemes without privacy preservation, the proposed approach increases the average computational and communication overhead across all parties by 2.09%.
Available online: April 29, 2026
, DOI: 10.13328/j.cnki.jos.007571
Abstract:
Access control technology is a security mechanism for managing users’ access to resources, which can effectively prevent unauthorized access and resource leakage. In the digital age, how to balance the relationship between information circulation and privacy protection via effective access control technology and ensure the safe and orderly flow of data elements has become a problem to be urgently addressed at present. However, the existing access control technology still has problems such as insufficient integration with trust evaluation, lack of dynamic adjustment capabilities, and difficulty in precise authorization in the research on domain data sharing scenarios. To this end, a trustworthy dynamic access control model scheme based on game theory is proposed, which integrates a three-layer collaborative mechanism of “trustworthy evaluation, dynamic adjustment, and access decision”. Firstly, based on the attribute weight algorithm, a multi-factor trustworthy prediction model is designed to calculate the trust probability of the access subject. Secondly, from the perspective of long-term stability, an evolutionary game dynamic adjustment model between the access subject and the object is built to periodically and dynamically adjust the reward and punishment incentive mechanism and access authorization threshold, thereby achieving the adaptive optimization of access control. Finally, based on Bayesian game theory, an incomplete information real-time decision-making model is built, and access control decisions are made based on the mixed strategy Nash equilibrium, with the trust degree updated by the equilibrium state feedback. The results of simulation experiments and sensitivity analysis verify that the proposed scheme can effectively improve the access control accuracy and achieve dynamic adjustment of access control strategies and accurate authorization.
Access control technology is a security mechanism for managing users’ access to resources, which can effectively prevent unauthorized access and resource leakage. In the digital age, how to balance the relationship between information circulation and privacy protection via effective access control technology and ensure the safe and orderly flow of data elements has become a problem to be urgently addressed at present. However, the existing access control technology still has problems such as insufficient integration with trust evaluation, lack of dynamic adjustment capabilities, and difficulty in precise authorization in the research on domain data sharing scenarios. To this end, a trustworthy dynamic access control model scheme based on game theory is proposed, which integrates a three-layer collaborative mechanism of “trustworthy evaluation, dynamic adjustment, and access decision”. Firstly, based on the attribute weight algorithm, a multi-factor trustworthy prediction model is designed to calculate the trust probability of the access subject. Secondly, from the perspective of long-term stability, an evolutionary game dynamic adjustment model between the access subject and the object is built to periodically and dynamically adjust the reward and punishment incentive mechanism and access authorization threshold, thereby achieving the adaptive optimization of access control. Finally, based on Bayesian game theory, an incomplete information real-time decision-making model is built, and access control decisions are made based on the mixed strategy Nash equilibrium, with the trust degree updated by the equilibrium state feedback. The results of simulation experiments and sensitivity analysis verify that the proposed scheme can effectively improve the access control accuracy and achieve dynamic adjustment of access control strategies and accurate authorization.
Available online: April 22, 2026
, DOI: 10.13328/j.cnki.jos.007615
Abstract:
With the rapid development of network technology, frequent cyber attacks, especially advanced persistent threats (APTs), seriously affect national security and social stability. Under the continuous evolution of encryption, obfuscation, and camouflage techniques, intelligent analysis of network traffic is considered an effective means to improve threat detection capability. However, when processing massive volumes of network traffic data, existing methods still suffer from high analysis complexity and weak model interpretability. Flow spectrum adopts domain transformation as a unified solution by constructing a more accurate, highly separable, and observable description space for network flow data, thereby enabling efficient characterization, representation and analysis of network behaviors and effectively addressing the above issues. Inspired by the atomic spectrum, this study proposes a novel flow spectrum scheme. The core idea is achieve a concrete representation of network behaviors by mapping network flows into a one-dimensional spectral space, and to detect network traffic threats through flow spectrum comparison, in which the design of an effective flow spectrum decomposer is crucial. In this study, the flow spectrum decomposer is constructed based on a semi-supervised autoencoder and is trained by jointly performing reconstruction and classification tasks, enabling spectral line distributions of different network behaviors to exhibit strong separability. The proposed scheme is validated on the NSL-KDD, UNSW-NB15, and CIC-DDoS2019 datasets. Experimental results show that the proposed scheme achieves high detection accuracy for network threat behaviors while providing differentiated representations for various network traffic behaviors, significantly enhancing network behavior observability and improving the interpretability of threat detection methods. Therefore, the proposed flow spectrum scheme is effective for network traffic threat detection.
With the rapid development of network technology, frequent cyber attacks, especially advanced persistent threats (APTs), seriously affect national security and social stability. Under the continuous evolution of encryption, obfuscation, and camouflage techniques, intelligent analysis of network traffic is considered an effective means to improve threat detection capability. However, when processing massive volumes of network traffic data, existing methods still suffer from high analysis complexity and weak model interpretability. Flow spectrum adopts domain transformation as a unified solution by constructing a more accurate, highly separable, and observable description space for network flow data, thereby enabling efficient characterization, representation and analysis of network behaviors and effectively addressing the above issues. Inspired by the atomic spectrum, this study proposes a novel flow spectrum scheme. The core idea is achieve a concrete representation of network behaviors by mapping network flows into a one-dimensional spectral space, and to detect network traffic threats through flow spectrum comparison, in which the design of an effective flow spectrum decomposer is crucial. In this study, the flow spectrum decomposer is constructed based on a semi-supervised autoencoder and is trained by jointly performing reconstruction and classification tasks, enabling spectral line distributions of different network behaviors to exhibit strong separability. The proposed scheme is validated on the NSL-KDD, UNSW-NB15, and CIC-DDoS2019 datasets. Experimental results show that the proposed scheme achieves high detection accuracy for network threat behaviors while providing differentiated representations for various network traffic behaviors, significantly enhancing network behavior observability and improving the interpretability of threat detection methods. Therefore, the proposed flow spectrum scheme is effective for network traffic threat detection.
Available online: April 22, 2026
, DOI: 10.13328/j.cnki.jos.007621
Abstract:
As the scale of open-source models and datasets continues to expand, the Hugging Face ecosystem forms a complex heterogeneous dependency network centered on model-dataset relationships. Issues such as missing metadata and high dependency concentration make chain-structured risks more likely to accumulate and propagate. To characterize this underlying risk landscape, this study constructs a resource dependency network of open-source AI resources based on a Hugging Face snapshot and analyzes its structural and evolutionary characteristics from the perspectives of global topology and temporal evolution. Furthermore, a “credibility risk” indicator is proposed by integrating metadata completeness and community feedback, enabling continuous risk quantification and ranking of model and dataset nodes. The results show that the dependency network exhibits a pronounced “spike-and-long-tail” structure, in which dependencies are highly concentrated on a small number of hub nodes, while a large proportion of resources remain isolated or semi-isolated within critical data-flow relationships. Meanwhile, the explosive growth of models, combined with the relatively slow expansion of datasets and contributors, strengthens the ecosystem’s path dependence on a limited set of core data sources, thereby giving rise to structural systemic risks. At the node level, the proposed credibility risk indicator demonstrates robustness to parameter perturbations and effectively distinguishes high-risk from low-risk nodes across multiple risk sources, outperforming baseline methods. Risk-coupling analysis and expert blind evaluation further confirm the clustering and propagation effects of high-risk datasets and high-risk models within local structures. Overall, this study provides a reproducible quantitative basis for risk screening and governance in open-source AI ecosystems.
As the scale of open-source models and datasets continues to expand, the Hugging Face ecosystem forms a complex heterogeneous dependency network centered on model-dataset relationships. Issues such as missing metadata and high dependency concentration make chain-structured risks more likely to accumulate and propagate. To characterize this underlying risk landscape, this study constructs a resource dependency network of open-source AI resources based on a Hugging Face snapshot and analyzes its structural and evolutionary characteristics from the perspectives of global topology and temporal evolution. Furthermore, a “credibility risk” indicator is proposed by integrating metadata completeness and community feedback, enabling continuous risk quantification and ranking of model and dataset nodes. The results show that the dependency network exhibits a pronounced “spike-and-long-tail” structure, in which dependencies are highly concentrated on a small number of hub nodes, while a large proportion of resources remain isolated or semi-isolated within critical data-flow relationships. Meanwhile, the explosive growth of models, combined with the relatively slow expansion of datasets and contributors, strengthens the ecosystem’s path dependence on a limited set of core data sources, thereby giving rise to structural systemic risks. At the node level, the proposed credibility risk indicator demonstrates robustness to parameter perturbations and effectively distinguishes high-risk from low-risk nodes across multiple risk sources, outperforming baseline methods. Risk-coupling analysis and expert blind evaluation further confirm the clustering and propagation effects of high-risk datasets and high-risk models within local structures. Overall, this study provides a reproducible quantitative basis for risk screening and governance in open-source AI ecosystems.
Available online: April 22, 2026
, DOI: 10.13328/j.cnki.jos.007626
Abstract:
In real-world scenarios such as e-commerce platforms and smart-home systems, multiple users often share the same account, resulting in behavioral sequences that mix interactions from different users. Sequential recommendation in shared-account settings has recently attracted increasing research attention but still faces two major challenges: (1) interactions from different latent users are difficult to disentangle accurately, leading to biased representation learning; (2) reliance on static hyperparameters to estimate the number of latent users prevents adaptive determination of user cardinality, which may cause underfitting or introduce noise. To address these challenges, this study proposes PE-GAT (prompt-enhanced graph attention network), an adaptive sequential recommendation framework for shared-account settings. PE-GAT first employs a dynamically-weighted, density-based clustering algorithm to infer the number of latent users and construct user-level sequential graphs that explicitly separate heterogeneous user behaviors. Based on these graphs, a graph attention network pre-training module is employed to disentangle mixed user preferences and learn initial sequence representations. Inspired by the pre-training and prompt-tuning paradigm, a self-attention-based prompt-enhancement module is then designed to refine sequence embeddings via prompt templates during retraining. Finally, account-level representations are fused with prompt-enhanced sequence embeddings to generate personalized recommendations. Extensive experiments on two real-world shared-account datasets (HVIDEO-E and HVIDEO-V) show that PE-GAT outperforms 14 state-of-the-art baselines, achieving maximum improvements of 4.73% and 5.59% in MRR and Recall metrics, respectively.
In real-world scenarios such as e-commerce platforms and smart-home systems, multiple users often share the same account, resulting in behavioral sequences that mix interactions from different users. Sequential recommendation in shared-account settings has recently attracted increasing research attention but still faces two major challenges: (1) interactions from different latent users are difficult to disentangle accurately, leading to biased representation learning; (2) reliance on static hyperparameters to estimate the number of latent users prevents adaptive determination of user cardinality, which may cause underfitting or introduce noise. To address these challenges, this study proposes PE-GAT (prompt-enhanced graph attention network), an adaptive sequential recommendation framework for shared-account settings. PE-GAT first employs a dynamically-weighted, density-based clustering algorithm to infer the number of latent users and construct user-level sequential graphs that explicitly separate heterogeneous user behaviors. Based on these graphs, a graph attention network pre-training module is employed to disentangle mixed user preferences and learn initial sequence representations. Inspired by the pre-training and prompt-tuning paradigm, a self-attention-based prompt-enhancement module is then designed to refine sequence embeddings via prompt templates during retraining. Finally, account-level representations are fused with prompt-enhanced sequence embeddings to generate personalized recommendations. Extensive experiments on two real-world shared-account datasets (HVIDEO-E and HVIDEO-V) show that PE-GAT outperforms 14 state-of-the-art baselines, achieving maximum improvements of 4.73% and 5.59% in MRR and Recall metrics, respectively.
Available online: April 22, 2026
, DOI: 10.13328/j.cnki.jos.007624
Abstract:
To address critical challenges in user privacy and transmission data security in cross-domain environments, as well as practical requirements for user portability and lightweight nodes in cross-domain communication system, this study proposes a blockchain-based secure cross-domain data transmission scheme under highly adversarial conditions. The scheme achieves: 1) decentralized public key management and identity authentication; 2) portable system access for users; 3) trusted key negotiation among communication nodes; 4) key-leakage resistant cross-domain data transmission. The security of the proposed scheme is formally analyzed and proven. Furthermore, a system prototype is implemented to evaluate performance. Experimental results indicate that computation, storage, and communication overheads are all efficiently managed, demonstrating the scheme’s practicality and scalability.
To address critical challenges in user privacy and transmission data security in cross-domain environments, as well as practical requirements for user portability and lightweight nodes in cross-domain communication system, this study proposes a blockchain-based secure cross-domain data transmission scheme under highly adversarial conditions. The scheme achieves: 1) decentralized public key management and identity authentication; 2) portable system access for users; 3) trusted key negotiation among communication nodes; 4) key-leakage resistant cross-domain data transmission. The security of the proposed scheme is formally analyzed and proven. Furthermore, a system prototype is implemented to evaluate performance. Experimental results indicate that computation, storage, and communication overheads are all efficiently managed, demonstrating the scheme’s practicality and scalability.
Available online: April 22, 2026
, DOI: 10.13328/j.cnki.jos.007579
Abstract:
Graph self-supervised learning aims to acquire effective graph structural representations without manual annotations. Although graph contrastive learning (GCL) enables self-supervised training by generating label-preserving perturbed views and maximizing their similarity to the original view, existing methods predominantly adopt a globally uniform perturbation strategy, which neglects the heterogeneity of node roles in real-world networks. Such indiscriminate random corruption violates the label-invariance assumption. Empirical observations indicate that real-world networks generally exhibit a core-periphery bi-layered topological architecture: core nodes form highly interconnected information hubs, and destructive perturbations applied to them are likely to cause semantic distortion and label shifts. To address these limitations, this study proposes a core-periphery structure-aware graph contrastive learning framework, with the following innovations: (i) instead of global uniform perturbation paradigm, peripheral nodes are accurately identified via a core-periphery detection algorithm, and localized perturbations are applied to preserve the integrity of the core topology, thereby strictly satisfying the principle of label invariance; (ii) augmentation operations such as peripheral node deletion are designed to simulate the dynamic evolution of real-world networks, encouraging the model to capture topological stability and noise robustness; (iii) a core-periphery contrastive loss function is constructed, in which differentiated weights are assigned to nodes with varying structural importance during loss computation, effectively guiding the model to emphasize core information while suppressing potential negative interference from peripheral nodes. Extensive experiments on multiple benchmark datasets demonstrate that the proposed method consistently outperforms state-of-the-art models across various tasks.
Graph self-supervised learning aims to acquire effective graph structural representations without manual annotations. Although graph contrastive learning (GCL) enables self-supervised training by generating label-preserving perturbed views and maximizing their similarity to the original view, existing methods predominantly adopt a globally uniform perturbation strategy, which neglects the heterogeneity of node roles in real-world networks. Such indiscriminate random corruption violates the label-invariance assumption. Empirical observations indicate that real-world networks generally exhibit a core-periphery bi-layered topological architecture: core nodes form highly interconnected information hubs, and destructive perturbations applied to them are likely to cause semantic distortion and label shifts. To address these limitations, this study proposes a core-periphery structure-aware graph contrastive learning framework, with the following innovations: (i) instead of global uniform perturbation paradigm, peripheral nodes are accurately identified via a core-periphery detection algorithm, and localized perturbations are applied to preserve the integrity of the core topology, thereby strictly satisfying the principle of label invariance; (ii) augmentation operations such as peripheral node deletion are designed to simulate the dynamic evolution of real-world networks, encouraging the model to capture topological stability and noise robustness; (iii) a core-periphery contrastive loss function is constructed, in which differentiated weights are assigned to nodes with varying structural importance during loss computation, effectively guiding the model to emphasize core information while suppressing potential negative interference from peripheral nodes. Extensive experiments on multiple benchmark datasets demonstrate that the proposed method consistently outperforms state-of-the-art models across various tasks.
Available online: April 09, 2026
, DOI: 10.13328/j.cnki.jos.007614
Abstract:
TLS protocol plays a critical role in ensuring the privacy, integrity, and reliability of network communications. In recent years, both industry and academia have actively promoted research and development related to TLS protocol, achieving significant progress, particularly in TLS 1.3. However, with the increasing complexity of network environments and the continuous evolution of attack methods, the security of the TLS 1.3 protocol also faces severe challenges, including replay attacks, risks to forward secrecy and vulnerabilities in software implementations such as OpenSSL. These attacks and vulnerabilities not only pose serious threats to user privacy and enterprise data security but also profoundly impact the trust system of the Internet and the broader digital economy. This study first provides a detailed introduction to the development and underlying principles of TLS 1.3. Subsequently, recent security research on the TLS 1.3 protocol is systematically categorized, compared, and analyzed from three aspects: protocol mechanism, software implementation, and application configuration. Finally, the current state and challenges in the TLS 1.3 protocol security research are summarized, and suggestions for future research directions are provided.
TLS protocol plays a critical role in ensuring the privacy, integrity, and reliability of network communications. In recent years, both industry and academia have actively promoted research and development related to TLS protocol, achieving significant progress, particularly in TLS 1.3. However, with the increasing complexity of network environments and the continuous evolution of attack methods, the security of the TLS 1.3 protocol also faces severe challenges, including replay attacks, risks to forward secrecy and vulnerabilities in software implementations such as OpenSSL. These attacks and vulnerabilities not only pose serious threats to user privacy and enterprise data security but also profoundly impact the trust system of the Internet and the broader digital economy. This study first provides a detailed introduction to the development and underlying principles of TLS 1.3. Subsequently, recent security research on the TLS 1.3 protocol is systematically categorized, compared, and analyzed from three aspects: protocol mechanism, software implementation, and application configuration. Finally, the current state and challenges in the TLS 1.3 protocol security research are summarized, and suggestions for future research directions are provided.
CHEN Miao, ZHANG Bi-Cheng, ZHANG Chen-Xi, PENG Xin, YANG Ding-Yu, LI Wei, QIAN Ze-Lin, WU Zhe-Shun, OU Jia-Yu, ZHONG Jian-Rui
Available online: April 01, 2026
, DOI: 10.13328/j.cnki.jos.007612
Abstract:
In large-scale online service systems, intricate dependencies among components often cause a single fault to trigger a massive number of correlated alerts, resulting in alert storms. Alert storms not only increase the workload of on-call engineers but also make fault diagnosis and root cause analysis more challenging. To address this issue, this study proposes Alert-CM, a graph representation learning-based alert clustering method that effectively groups alerts caused by the same fault, thereby alleviating engineers’ workloads. In alert management, an alert is typically generated by aggregating various types of underlying system data at the time of a fault, including related applications, metrics, logs, alert rules, and emergency scenarios. Alert-CM assumes that multiple alerts triggered by the same fault usually exhibit tight correlations at the underlying system data level, and that the core underlying system data better represents the abnormal system state associated with the alerts. Based on this assumption, Alert-CM constructs a fine-grained system data dependency graph using alert-related configuration data, abstracting and mapping the dependency relationships between alerts and graph nodes to further expand the alert feature space. On top of the system data dependency graph, a graph neural network model is built to perform graph representation learning, in which the contributions of core underlying system data to alerts are automatically learned, producing accurate alert vector representations. Finally, the DBSCAN algorithm is applied to cluster alerts based on the learned representations. Alert-CM is evaluated on a real-world industrial dataset, with a focus on clustering effectiveness and real-time efficiency. The experimental results demonstrate that Alert-CM significantly outperforms traditional alert aggregation methods in alert clustering tasks. Specifically, Alert-CM achieves an NMI of 0.901 and an ARI of 0.645, corresponding to average improvements of 31.7% and 153.9% over existing methods, respectively. In addition, Alert-CM exhibits strong performance in online real-time clustering tasks.
In large-scale online service systems, intricate dependencies among components often cause a single fault to trigger a massive number of correlated alerts, resulting in alert storms. Alert storms not only increase the workload of on-call engineers but also make fault diagnosis and root cause analysis more challenging. To address this issue, this study proposes Alert-CM, a graph representation learning-based alert clustering method that effectively groups alerts caused by the same fault, thereby alleviating engineers’ workloads. In alert management, an alert is typically generated by aggregating various types of underlying system data at the time of a fault, including related applications, metrics, logs, alert rules, and emergency scenarios. Alert-CM assumes that multiple alerts triggered by the same fault usually exhibit tight correlations at the underlying system data level, and that the core underlying system data better represents the abnormal system state associated with the alerts. Based on this assumption, Alert-CM constructs a fine-grained system data dependency graph using alert-related configuration data, abstracting and mapping the dependency relationships between alerts and graph nodes to further expand the alert feature space. On top of the system data dependency graph, a graph neural network model is built to perform graph representation learning, in which the contributions of core underlying system data to alerts are automatically learned, producing accurate alert vector representations. Finally, the DBSCAN algorithm is applied to cluster alerts based on the learned representations. Alert-CM is evaluated on a real-world industrial dataset, with a focus on clustering effectiveness and real-time efficiency. The experimental results demonstrate that Alert-CM significantly outperforms traditional alert aggregation methods in alert clustering tasks. Specifically, Alert-CM achieves an NMI of 0.901 and an ARI of 0.645, corresponding to average improvements of 31.7% and 153.9% over existing methods, respectively. In addition, Alert-CM exhibits strong performance in online real-time clustering tasks.
Available online: April 01, 2026
, DOI: 10.13328/j.cnki.jos.007613
Abstract:
Large language models (LLMs) demonstrate significant potential in automatic code generation. However, in practical applications, the generated code often suffers from multiple issues, including syntax errors, semantics inconsistencies, security inconsistencies, inefficient runtime performance, and poor maintainability. To address these challenges, constrained code generation techniques are introduced. Drawing inspiration from constrained text generation, these techniques impose explicit constraints at various stages of the code generation process to ensure that the generated code satisfies predefined requirements. This study first reviews the major issues exposed by LLMs in code generation, with a detailed analysis of deficiencies related to code correctness and quality. Subsequently, recent research progress in constrained code generation is summarized, and the strengths and limitations of existing approaches are systematically examined. Furthermore, evaluation methods are discussed, including the construction of benchmark datasets and the design of evaluation metrics, providing valuable references for experimental settings in future research. Finally, the research challenges faced by constrained code generation and its future development trends are outlined.
Large language models (LLMs) demonstrate significant potential in automatic code generation. However, in practical applications, the generated code often suffers from multiple issues, including syntax errors, semantics inconsistencies, security inconsistencies, inefficient runtime performance, and poor maintainability. To address these challenges, constrained code generation techniques are introduced. Drawing inspiration from constrained text generation, these techniques impose explicit constraints at various stages of the code generation process to ensure that the generated code satisfies predefined requirements. This study first reviews the major issues exposed by LLMs in code generation, with a detailed analysis of deficiencies related to code correctness and quality. Subsequently, recent research progress in constrained code generation is summarized, and the strengths and limitations of existing approaches are systematically examined. Furthermore, evaluation methods are discussed, including the construction of benchmark datasets and the design of evaluation metrics, providing valuable references for experimental settings in future research. Finally, the research challenges faced by constrained code generation and its future development trends are outlined.
Available online: March 25, 2026
, DOI: 10.13328/j.cnki.jos.007568
Abstract:
As a key technology for ensuring the safety and reliability of artificial intelligence (AI) systems, neural network robustness verification can provide formal guarantees for intelligent decision-making. Existing research is generally based on simplified assumptions of isotropic data distributions to develop verification algorithms based on uniform Lp-norm ball neighborhoods. However, this theoretical framework proves inadequate in the face of the real-world complex data characteristics. For instance, different data features exhibit varying influences on model predictions and sensitivities to perturbations; some features are immutable due to physical constraints; complex correlation structures may exist among features. This makes it difficult for verification algorithms based on uniform perturbation domains to accurately model the robustness requirements of the real world for AI systems. To this end, this study proposes a robustness verification framework based on non-uniform perturbation domains. By combining the data distribution characteristics of specific application domains, the study constructs geometrically structured perturbation domains aligned with domain-specific features and model domain-adaptive perturbations. On this basis, it formally defines three novel robustness concepts, including ellipsoidal robustness, masked local robustness, and Mahalanobis distance robustness, and proposes the definition and construction methods for corresponding robustness verification problems. Furthermore, the NNV4RADAP algorithm is designed, which extends existing verification algorithms to neural network robustness verification problems for domain-adaptive perturbations by constructing equivalent uniform Lp-norm ball robustness verification problems. The experimental results demonstrate that the NNV4RADAP algorithm can provide more accurate and datadistribution-aligned robustness guarantees for neural networks. This study expands the existing formal definitions of deep neural network robustness and designs and implements a formal verification algorithm of neural network robustness for domain-adaptive perturbations. Additionally, it researches the problem in data distribution-based robustness definitions and provides guidance for the future implementation and application of formal verification techniques in trustworthy AI technologies.
As a key technology for ensuring the safety and reliability of artificial intelligence (AI) systems, neural network robustness verification can provide formal guarantees for intelligent decision-making. Existing research is generally based on simplified assumptions of isotropic data distributions to develop verification algorithms based on uniform Lp-norm ball neighborhoods. However, this theoretical framework proves inadequate in the face of the real-world complex data characteristics. For instance, different data features exhibit varying influences on model predictions and sensitivities to perturbations; some features are immutable due to physical constraints; complex correlation structures may exist among features. This makes it difficult for verification algorithms based on uniform perturbation domains to accurately model the robustness requirements of the real world for AI systems. To this end, this study proposes a robustness verification framework based on non-uniform perturbation domains. By combining the data distribution characteristics of specific application domains, the study constructs geometrically structured perturbation domains aligned with domain-specific features and model domain-adaptive perturbations. On this basis, it formally defines three novel robustness concepts, including ellipsoidal robustness, masked local robustness, and Mahalanobis distance robustness, and proposes the definition and construction methods for corresponding robustness verification problems. Furthermore, the NNV4RADAP algorithm is designed, which extends existing verification algorithms to neural network robustness verification problems for domain-adaptive perturbations by constructing equivalent uniform Lp-norm ball robustness verification problems. The experimental results demonstrate that the NNV4RADAP algorithm can provide more accurate and datadistribution-aligned robustness guarantees for neural networks. This study expands the existing formal definitions of deep neural network robustness and designs and implements a formal verification algorithm of neural network robustness for domain-adaptive perturbations. Additionally, it researches the problem in data distribution-based robustness definitions and provides guidance for the future implementation and application of formal verification techniques in trustworthy AI technologies.
Available online: March 25, 2026
, DOI: 10.13328/j.cnki.jos.007569
Abstract:
Image denoising, a fundamental task in computer vision, relies on leveraging effective prior knowledge to restore detailed information corrupted by noise. This study proposes a feature spatial contextual Transformer for image denoising FSCformer to address the performance bottlenecks of traditional convolutional neural network (CNN) caused by fixed weights and the limitation of local receptive fields, and the high computational complexity of Transformer for global modeling. In this method, an efficient receptive field module is introduced to dynamically capture multi-scale contextual information, enhancing spatial perception while significantly reducing computational overhead. Furthermore, a convolutional attention module is adopted to integrate local feature extraction with global dependency modeling, improving the robustness of the model especially in complex scenes. Additionally, a cross-feature fusion mechanism is proposed to preserve image details by promoting the fine-grained interaction of multi-scale features. Extensive experiments demonstrate that the proposed method achieves a favorable trade-off between denoising accuracy and computational efficiency, outperforming several existing image denoising models across multiple benchmark datasets.
Image denoising, a fundamental task in computer vision, relies on leveraging effective prior knowledge to restore detailed information corrupted by noise. This study proposes a feature spatial contextual Transformer for image denoising FSCformer to address the performance bottlenecks of traditional convolutional neural network (CNN) caused by fixed weights and the limitation of local receptive fields, and the high computational complexity of Transformer for global modeling. In this method, an efficient receptive field module is introduced to dynamically capture multi-scale contextual information, enhancing spatial perception while significantly reducing computational overhead. Furthermore, a convolutional attention module is adopted to integrate local feature extraction with global dependency modeling, improving the robustness of the model especially in complex scenes. Additionally, a cross-feature fusion mechanism is proposed to preserve image details by promoting the fine-grained interaction of multi-scale features. Extensive experiments demonstrate that the proposed method achieves a favorable trade-off between denoising accuracy and computational efficiency, outperforming several existing image denoising models across multiple benchmark datasets.
Available online: March 25, 2026
, DOI: 10.13328/j.cnki.jos.007643
Abstract:
Pretrained code models have demonstrated strong capabilities in code understanding and analysis and have become important tools and major research focuses in source code vulnerability detection. However, similar to traditional deep learning models, pretrained code models exhibit robustness vulnerabilities when exposed to carefully crafted adversarial code. Attackers can mislead the model into classifying vulnerable code as non-vulnerable by introducing semantically preserving perturbations, thus posing a significant threat to software security. Therefore, adversarial attacks for pretrained vulnerability detection models not only serve as an effective approach to evaluate the robustness of pretrained code models but also provide critical insights for the development of future vulnerability detection models and defense mechanisms. In the hard-label black-box attack scenario, this study proposes VulBlurrer, a black-box adversarial attack method targeting pretrained vulnerability detection models. VulBlurrer designs a directed synonymous code transformation strategy, prioritizing perturbations on highly sensitive regions adjacent to vulnerabilities and on specific statements. It also introduces an escape score based on feature consistency, semantic consistency, and code fluency, enabling the quantification of the potential attack value of candidate samples without accessing internal information of the target model. Furthermore, this study adopts a genetic-algorithm-based optimization strategy for adversarial code, in which the weights used to compute the escape score are dynamically adjusted and an elite retention mechanism is applied during the iterative process, thus further improving attack accuracy. VulBlurrer and baseline methods are evaluated on pretrained vulnerability detection models based on CodeBERT, GraphCodeBERT, CodeT5, and UniXcoder. The attack success rates of VulBlurrer on the four target models reach 85.51%, 91.47%, 93.14%, and 71.61%, with average query numbers of 12.67, 9.10, 11.07, and 19.44, respectively. Compared with existing methods, VulBlurrer achieves higher attack success rates and a better trade-off between attack success rate and query efficiency, while the generated adversarial code also exhibit superior consistency and fluency. Furthermore, empirical studies are conducted on ChatGPT, DeepSeek, and LLM-based programming assistants, including GitHub Copilot and TRAE, verifying the effectiveness of the proposed method against large language models. These results indicate that pretrained code models still face robustness challenges posed by adversarial attacks in vulnerability detection tasks, and vulnerability detection tools based on pretrained models and large language models require further improvements to enhance robustness against adversarial code.
Pretrained code models have demonstrated strong capabilities in code understanding and analysis and have become important tools and major research focuses in source code vulnerability detection. However, similar to traditional deep learning models, pretrained code models exhibit robustness vulnerabilities when exposed to carefully crafted adversarial code. Attackers can mislead the model into classifying vulnerable code as non-vulnerable by introducing semantically preserving perturbations, thus posing a significant threat to software security. Therefore, adversarial attacks for pretrained vulnerability detection models not only serve as an effective approach to evaluate the robustness of pretrained code models but also provide critical insights for the development of future vulnerability detection models and defense mechanisms. In the hard-label black-box attack scenario, this study proposes VulBlurrer, a black-box adversarial attack method targeting pretrained vulnerability detection models. VulBlurrer designs a directed synonymous code transformation strategy, prioritizing perturbations on highly sensitive regions adjacent to vulnerabilities and on specific statements. It also introduces an escape score based on feature consistency, semantic consistency, and code fluency, enabling the quantification of the potential attack value of candidate samples without accessing internal information of the target model. Furthermore, this study adopts a genetic-algorithm-based optimization strategy for adversarial code, in which the weights used to compute the escape score are dynamically adjusted and an elite retention mechanism is applied during the iterative process, thus further improving attack accuracy. VulBlurrer and baseline methods are evaluated on pretrained vulnerability detection models based on CodeBERT, GraphCodeBERT, CodeT5, and UniXcoder. The attack success rates of VulBlurrer on the four target models reach 85.51%, 91.47%, 93.14%, and 71.61%, with average query numbers of 12.67, 9.10, 11.07, and 19.44, respectively. Compared with existing methods, VulBlurrer achieves higher attack success rates and a better trade-off between attack success rate and query efficiency, while the generated adversarial code also exhibit superior consistency and fluency. Furthermore, empirical studies are conducted on ChatGPT, DeepSeek, and LLM-based programming assistants, including GitHub Copilot and TRAE, verifying the effectiveness of the proposed method against large language models. These results indicate that pretrained code models still face robustness challenges posed by adversarial attacks in vulnerability detection tasks, and vulnerability detection tools based on pretrained models and large language models require further improvements to enhance robustness against adversarial code.
Available online: February 11, 2026
, DOI: 10.13328/j.cnki.jos.007566
Abstract:
As autonomous driving applications are rapidly popularized, their safety has become the common focus of both academia and industry. Autonomous driving system (ADS) testing is an effective means for solving this problem. Currently, the mainstream testing method is the scenario-based simulation test, which evaluates the decision of ADS to be measured by simulating various elements of driving scenarios, such as roads and pedestrians. However, existing methods mainly focus on the construction and dynamic generation of critical driving scenarios, neglecting the influence of configuration changes of the vehicle itself, such as its weight and torque, on the decision-making of ADS deployed on the vehicle. To address this issue, based on the previous work SAFEVAR, this study proposes SAFEVCS, an efficient search method for safety-critical vehicle configurations. SAFEVCS employs a search algorithm to explore the vehicle configuration setting (VCS) that exposes safety vulnerabilities of ADS. Furthermore, to improve the diversity of the search results, SAFEVCS introduces fuzzing to optimize the conditions and constraints of crossover and mutation operators in search algorithms. To improve search efficiency, SAFEVCS further combines the vehicle dynamics knowledge, which achieves the self-adaption of search termination strategy and deduplication strategy. To evaluate the effectiveness and execution efficiency of SAFEVCS, the study takes SAFEVAR as the baseline for comparison and carries out extensive experiments under three driving scenarios. The experimental results show that VCS generated by SAFEVCS can effectively expose the safety vulnerabilities of ADS. In the two weather conditions of sunny and rainy days, under the simulation scenarios of pedestrians crossing the road, the obtained solution set significantly decreased the safety performance of the ADS under test, and under the same experiment environment, the simulation efficiency is increased by approximately 2.5 times.
As autonomous driving applications are rapidly popularized, their safety has become the common focus of both academia and industry. Autonomous driving system (ADS) testing is an effective means for solving this problem. Currently, the mainstream testing method is the scenario-based simulation test, which evaluates the decision of ADS to be measured by simulating various elements of driving scenarios, such as roads and pedestrians. However, existing methods mainly focus on the construction and dynamic generation of critical driving scenarios, neglecting the influence of configuration changes of the vehicle itself, such as its weight and torque, on the decision-making of ADS deployed on the vehicle. To address this issue, based on the previous work SAFEVAR, this study proposes SAFEVCS, an efficient search method for safety-critical vehicle configurations. SAFEVCS employs a search algorithm to explore the vehicle configuration setting (VCS) that exposes safety vulnerabilities of ADS. Furthermore, to improve the diversity of the search results, SAFEVCS introduces fuzzing to optimize the conditions and constraints of crossover and mutation operators in search algorithms. To improve search efficiency, SAFEVCS further combines the vehicle dynamics knowledge, which achieves the self-adaption of search termination strategy and deduplication strategy. To evaluate the effectiveness and execution efficiency of SAFEVCS, the study takes SAFEVAR as the baseline for comparison and carries out extensive experiments under three driving scenarios. The experimental results show that VCS generated by SAFEVCS can effectively expose the safety vulnerabilities of ADS. In the two weather conditions of sunny and rainy days, under the simulation scenarios of pedestrians crossing the road, the obtained solution set significantly decreased the safety performance of the ADS under test, and under the same experiment environment, the simulation efficiency is increased by approximately 2.5 times.
Available online: February 04, 2026
, DOI: 10.13328/j.cnki.jos.007572
Abstract:
As a superset of JavaScript, TypeScript provides a rich set of features, such as static type support and object-oriented programming capabilities. It is widely adopted by many mainstream frameworks such as Angular, Vue, and React, and has become a core technology for building large-scale applications. Its compiler is responsible for compiling TypeScript codes into standard JavaScript codes. However, the TypeScript compiler itself may contain bugs, resulting in unexpected errors in the generated JavaScript code. Although numerous studies have been conducted on JavaScript engine testing, there has been no systematic study dedicated to testing the TypeScript compiler. Existing JavaScript engine testing methods have difficulty in generating a large number of TypeScript programs with specific types and effectively mutating these types, thus making it difficult to detect bugs related to complex type systems in the TypeScript compiler. To this end, a TypeScript compiler testing framework based on syntax and type mutation TscFuzz is proposed. To obtain a large number of seed programs containing specific types of TypeScript, TscFuzz designs a set of prompts tailored to the unique type system of TypeScript compared to JavaScript, with the large language model (LLM) guided to generate a series of programs featuring these specific types. Next, a set of type-specific mutation operators are designed to conduct targeted testing on the type system of TypeScript via type mutation. Finally, based on differential testing of the cross-version strategy, TscFuzz compares the outputs of different versions of the TypeScript compiler to detect bugs. Additionally, Node.js is employed to verify the semantic correctness of the JavaScript programs output by the compiler. Experimental results demonstrate that TscFuzz detects five bugs within 72 hours, two and three bugs more than the baseline methods DIE and FuzzJIT, respectively. The bug detection effect of TscFuzz is significantly better than that of the baseline methods. Meanwhile, after three months of testing, TscFuzz successfully identifies 12 real TypeScript bugs, eight of which have been confirmed and seven have been repaired.
As a superset of JavaScript, TypeScript provides a rich set of features, such as static type support and object-oriented programming capabilities. It is widely adopted by many mainstream frameworks such as Angular, Vue, and React, and has become a core technology for building large-scale applications. Its compiler is responsible for compiling TypeScript codes into standard JavaScript codes. However, the TypeScript compiler itself may contain bugs, resulting in unexpected errors in the generated JavaScript code. Although numerous studies have been conducted on JavaScript engine testing, there has been no systematic study dedicated to testing the TypeScript compiler. Existing JavaScript engine testing methods have difficulty in generating a large number of TypeScript programs with specific types and effectively mutating these types, thus making it difficult to detect bugs related to complex type systems in the TypeScript compiler. To this end, a TypeScript compiler testing framework based on syntax and type mutation TscFuzz is proposed. To obtain a large number of seed programs containing specific types of TypeScript, TscFuzz designs a set of prompts tailored to the unique type system of TypeScript compared to JavaScript, with the large language model (LLM) guided to generate a series of programs featuring these specific types. Next, a set of type-specific mutation operators are designed to conduct targeted testing on the type system of TypeScript via type mutation. Finally, based on differential testing of the cross-version strategy, TscFuzz compares the outputs of different versions of the TypeScript compiler to detect bugs. Additionally, Node.js is employed to verify the semantic correctness of the JavaScript programs output by the compiler. Experimental results demonstrate that TscFuzz detects five bugs within 72 hours, two and three bugs more than the baseline methods DIE and FuzzJIT, respectively. The bug detection effect of TscFuzz is significantly better than that of the baseline methods. Meanwhile, after three months of testing, TscFuzz successfully identifies 12 real TypeScript bugs, eight of which have been confirmed and seven have been repaired.
Available online: February 04, 2026
, DOI: 10.13328/j.cnki.jos.007575
Abstract:
As a widely employed interpreted language, Python faces performance challenges in execution efficiency. Just-in-time (JIT) compilers have been introduced to the Python ecosystem to dynamically compile bytecode into machine code, significantly improving program operation speed. However, the complex optimization strategies of JIT compilers may introduce program defects, thereby affecting program stability and reliability. Existing fuzz testing methods for Python interpreters struggle to effectively detect deep optimization defects and non-crashing defects in JIT compilers. To this end, this study proposes PjitFuzz, a coverage-guided defect detection method for Python JIT compilers. First, PjitFuzz proposes five mutation rules based on JIT optimization strategies to generate program variants that trigger the optimization strategies of Python JIT compilers. Second, a coverage-guided dynamic mutation rule selection method is designed to integrate the advantages of different mutation rules and generate diverse program variants. Third, a checksum-based code block insertion strategy is developed to effectively record changes in variable values during program execution and detect inconsistency in the output. Finally, differential testing is performed by combining different JIT compilation options to effectively detect defects in Python JIT compilers. This study compares PjitFuzz with two state-of-the-art Python interpreter fuzzing methods, FcFuzzer and IFuzzer. The experimental results show that PjitFuzz improves defect detection capability by 150% and 66.7% respectively, and outperforms existing methods in terms of code coverage by 28.23% and 15.68% respectively. For the validity rate of generated test programs, PjitFuzz outperforms the comparative methods by 42.42% and 62.74% respectively. In an eight-month experiment, PjitFuzz has discovered and reported 16 defects, 12 of which have been confirmed by developers.
As a widely employed interpreted language, Python faces performance challenges in execution efficiency. Just-in-time (JIT) compilers have been introduced to the Python ecosystem to dynamically compile bytecode into machine code, significantly improving program operation speed. However, the complex optimization strategies of JIT compilers may introduce program defects, thereby affecting program stability and reliability. Existing fuzz testing methods for Python interpreters struggle to effectively detect deep optimization defects and non-crashing defects in JIT compilers. To this end, this study proposes PjitFuzz, a coverage-guided defect detection method for Python JIT compilers. First, PjitFuzz proposes five mutation rules based on JIT optimization strategies to generate program variants that trigger the optimization strategies of Python JIT compilers. Second, a coverage-guided dynamic mutation rule selection method is designed to integrate the advantages of different mutation rules and generate diverse program variants. Third, a checksum-based code block insertion strategy is developed to effectively record changes in variable values during program execution and detect inconsistency in the output. Finally, differential testing is performed by combining different JIT compilation options to effectively detect defects in Python JIT compilers. This study compares PjitFuzz with two state-of-the-art Python interpreter fuzzing methods, FcFuzzer and IFuzzer. The experimental results show that PjitFuzz improves defect detection capability by 150% and 66.7% respectively, and outperforms existing methods in terms of code coverage by 28.23% and 15.68% respectively. For the validity rate of generated test programs, PjitFuzz outperforms the comparative methods by 42.42% and 62.74% respectively. In an eight-month experiment, PjitFuzz has discovered and reported 16 defects, 12 of which have been confirmed by developers.
Available online: February 04, 2026
, DOI: 10.13328/j.cnki.jos.007577
Abstract:
In recent years, recommender systems based on graph neural network (GNN) have made good use of the interaction structure of interaction data to learn user and item representations. However, existing recommendation models based on GNN often ignore the temporal information of interactions during aggregation, which makes it difficult to model the change characteristics of users’ interests. As a result, this causes overfitting of the recommendation model to data, and a lack of diversity in the recommendation results, thereby making it difficult to satisfy the more diversified needs of users. To this end, a temporal information-enhanced diversified recommendation model is proposed. First, an attention mechanism is employed to capture and fuse temporal information and interaction information from historical user-item interactions. Meanwhile, a feature disentanglement module is designed to disentangle smoothed global features from salient, highly discriminative key signals to reduce feature redundancy and improve representational clarity. Subsequently, neighbour selection is adopted to highlight inter-node differences and conduct graph convolution, with a layer attention mechanism employed to alleviate over smoothing. Finally, the learning of items in the long-tail category is enhanced by reweighting loss to improve the diversity.
In recent years, recommender systems based on graph neural network (GNN) have made good use of the interaction structure of interaction data to learn user and item representations. However, existing recommendation models based on GNN often ignore the temporal information of interactions during aggregation, which makes it difficult to model the change characteristics of users’ interests. As a result, this causes overfitting of the recommendation model to data, and a lack of diversity in the recommendation results, thereby making it difficult to satisfy the more diversified needs of users. To this end, a temporal information-enhanced diversified recommendation model is proposed. First, an attention mechanism is employed to capture and fuse temporal information and interaction information from historical user-item interactions. Meanwhile, a feature disentanglement module is designed to disentangle smoothed global features from salient, highly discriminative key signals to reduce feature redundancy and improve representational clarity. Subsequently, neighbour selection is adopted to highlight inter-node differences and conduct graph convolution, with a layer attention mechanism employed to alleviate over smoothing. Finally, the learning of items in the long-tail category is enhanced by reweighting loss to improve the diversity.
Available online: February 04, 2026
, DOI: 10.13328/j.cnki.jos.007578
Abstract:
As multimodal multiobjective optimization faces challenges of reasonably defining the individual crowdedness and dynamically balancing the decision space and objective space in individual diversity calculation, there is still significant room for performance improvement in existing multimodal multiobjective optimization algorithms. To this end, this study proposes a multimodal multiobjective differential evolution algorithm based on adaptive individual diversity (MMODE-AID). First, based on the average Euclidean distance of individuals’ nearest neighbors in the decision space or objective space, the crowdedness of individuals can be defined by multiplying the relative distances between individuals, which can more reasonably measure the true crowdedness of each individual in the corresponding space. Second, based on the overall crowdedness of the decision space and objective space, the relative crowdedness of individuals in the corresponding space is obtained, which can reasonably and dynamically balance the influence of the current state of the decision space and objective space on individual diversity calculation during the evolution process, and is conducive to the sufficient search of each equivalent Pareto optimal solution set. By employing differential evolution as the basic optimization framework, MMODE-AID evaluates individual fitness based on adaptive individual diversity. Meanwhile, it can obtain a population with excellent performance in decision space distribution, objective space distribution and convergence during offspring generation and environmental selection. MMODE-AID is compared with seven advanced multimodal multiobjective optimization algorithms on 39 benchmark test problems and one real-world application problem to validate the algorithm’s performance. The experimental results demonstrate that MMODE-AID exhibits significant competitive advantages in solving multimodal multiobjective optimization problems. The source code and original experimental data of MMODE-AID are publicly available on GitHub: https://github.com/CIA-SZU/ZQ.
As multimodal multiobjective optimization faces challenges of reasonably defining the individual crowdedness and dynamically balancing the decision space and objective space in individual diversity calculation, there is still significant room for performance improvement in existing multimodal multiobjective optimization algorithms. To this end, this study proposes a multimodal multiobjective differential evolution algorithm based on adaptive individual diversity (MMODE-AID). First, based on the average Euclidean distance of individuals’ nearest neighbors in the decision space or objective space, the crowdedness of individuals can be defined by multiplying the relative distances between individuals, which can more reasonably measure the true crowdedness of each individual in the corresponding space. Second, based on the overall crowdedness of the decision space and objective space, the relative crowdedness of individuals in the corresponding space is obtained, which can reasonably and dynamically balance the influence of the current state of the decision space and objective space on individual diversity calculation during the evolution process, and is conducive to the sufficient search of each equivalent Pareto optimal solution set. By employing differential evolution as the basic optimization framework, MMODE-AID evaluates individual fitness based on adaptive individual diversity. Meanwhile, it can obtain a population with excellent performance in decision space distribution, objective space distribution and convergence during offspring generation and environmental selection. MMODE-AID is compared with seven advanced multimodal multiobjective optimization algorithms on 39 benchmark test problems and one real-world application problem to validate the algorithm’s performance. The experimental results demonstrate that MMODE-AID exhibits significant competitive advantages in solving multimodal multiobjective optimization problems. The source code and original experimental data of MMODE-AID are publicly available on GitHub: https://github.com/CIA-SZU/ZQ.
Available online: January 21, 2026
, DOI: 10.13328/j.cnki.jos.007562
Abstract:
The current software market is witnessing an intensified trend of product homogenization, where functional innovation has become a decisive factor in maintaining competitive advantage. This shift has transformed the paradigm of modern requirements engineering from passive requirements extraction to proactive creative requirements capture. Existing approaches to enhancing requirements creativity primarily follow two paths: (1) fostering collaborative innovation in workshops through scenario modeling and facilitation methods, and (2) rapidly generating novel solutions by deconstructing and recombining existing requirements based on combinatorial innovation theory. However, both methods face a core challenge in balancing innovation quality with participation costs. The breakthrough advancements in generative AI technologies offer new opportunities to address this dilemma. This study proposes a business modeling-driven human-AI multi-agent collaborative framework with TRIZ infusion for creative requirements capture (BMHACT). The framework adopts the unified process business modeling collaborative architecture to design prompt-based definitions for five agent roles: business process analyst, business designer, and other relevant roles. The multi-agent team collaboratively generates creative requirements through a structured workflow: system vision collection→process pain point identification→technical contradiction analysis→TRIZ innovation principle matching→requirement solution generation. Domain experts and client representatives then evaluate the requirements for creativity. An empirical study on a portal system for a small-scale mechanical manufacturing enterprise demonstrates that, compared to the requirement reuse-based method and the adversarial-sample-based retrospective requirement generation method, BMHACT reduces iteration cycles by 50% and 28.6%, shortens total process duration by 66.7% and 33.3%, increases the clarity novelty usefulness (CNU) by 22.9% and 10.7%, and achieves a 2.16× and 2.14× higher per-round CNU improvement rate. These results validate BMHACT’s superiority in enhancing requirements innovation quality while reducing collaboration costs.
The current software market is witnessing an intensified trend of product homogenization, where functional innovation has become a decisive factor in maintaining competitive advantage. This shift has transformed the paradigm of modern requirements engineering from passive requirements extraction to proactive creative requirements capture. Existing approaches to enhancing requirements creativity primarily follow two paths: (1) fostering collaborative innovation in workshops through scenario modeling and facilitation methods, and (2) rapidly generating novel solutions by deconstructing and recombining existing requirements based on combinatorial innovation theory. However, both methods face a core challenge in balancing innovation quality with participation costs. The breakthrough advancements in generative AI technologies offer new opportunities to address this dilemma. This study proposes a business modeling-driven human-AI multi-agent collaborative framework with TRIZ infusion for creative requirements capture (BMHACT). The framework adopts the unified process business modeling collaborative architecture to design prompt-based definitions for five agent roles: business process analyst, business designer, and other relevant roles. The multi-agent team collaboratively generates creative requirements through a structured workflow: system vision collection→process pain point identification→technical contradiction analysis→TRIZ innovation principle matching→requirement solution generation. Domain experts and client representatives then evaluate the requirements for creativity. An empirical study on a portal system for a small-scale mechanical manufacturing enterprise demonstrates that, compared to the requirement reuse-based method and the adversarial-sample-based retrospective requirement generation method, BMHACT reduces iteration cycles by 50% and 28.6%, shortens total process duration by 66.7% and 33.3%, increases the clarity novelty usefulness (CNU) by 22.9% and 10.7%, and achieves a 2.16× and 2.14× higher per-round CNU improvement rate. These results validate BMHACT’s superiority in enhancing requirements innovation quality while reducing collaboration costs.
Available online: January 21, 2026
, DOI: 10.13328/j.cnki.jos.007472
Abstract:
The heuristic test case generation method that combines machine learning-related technologies can significantly improve the test efficiency. Existing studies focus on building efficient surrogate models with partial test cases, but ignore the influence of both the initial population quality and surrogate models on the multi-path testing efficiency. Therefore, this study proposes a test case reduction and generation method combining K-means and support vector machine regression (SVR). The randomly generated test cases are clustered into several clusters by adopting K-means, and only the test cases that are within a particular distance away from the cluster center are retained, with the path coverage matrix for these test cases constructed. This matrix is employed to evaluate the path coverage potential of test cases and the coverage difficulty of paths. Additionally, based on these two conditions, the test cases are ranked, and several test cases are selected from different clusters to construct the test case reduction set, which is taken as the initial genetic population. This not only increases the diversity of the initial population and reduces its redundancy, but also helps to reduce the iteration number for multi-path coverage test cases. Meanwhile, the test cases before clustering and their fitness are employed as the samples to train the SVR fitness prediction model designed for multi-path coverage, and then the new test cases generated by genetic evolution are utilized to update the model, thus improving the model accuracy and reducing the time consumed due to the instrumentation program execution. In this way, both population quality and test efficiency can be improved. The experimental results show that on fifteen programs, the proposed method has better improvements in terms of indicators such as the coverage rate and average evolutionary generation. Specifically, in terms of the coverage rate, the proposed method demonstrates an improvement of at least 7% and up to 49% compared to three types of baseline methods, and shows the enhancement of approximately 10% to a maximum of 25% compared to five competitive methods. The proposed method provides guidance for the research on multi-path testing that combines machine learning.
The heuristic test case generation method that combines machine learning-related technologies can significantly improve the test efficiency. Existing studies focus on building efficient surrogate models with partial test cases, but ignore the influence of both the initial population quality and surrogate models on the multi-path testing efficiency. Therefore, this study proposes a test case reduction and generation method combining K-means and support vector machine regression (SVR). The randomly generated test cases are clustered into several clusters by adopting K-means, and only the test cases that are within a particular distance away from the cluster center are retained, with the path coverage matrix for these test cases constructed. This matrix is employed to evaluate the path coverage potential of test cases and the coverage difficulty of paths. Additionally, based on these two conditions, the test cases are ranked, and several test cases are selected from different clusters to construct the test case reduction set, which is taken as the initial genetic population. This not only increases the diversity of the initial population and reduces its redundancy, but also helps to reduce the iteration number for multi-path coverage test cases. Meanwhile, the test cases before clustering and their fitness are employed as the samples to train the SVR fitness prediction model designed for multi-path coverage, and then the new test cases generated by genetic evolution are utilized to update the model, thus improving the model accuracy and reducing the time consumed due to the instrumentation program execution. In this way, both population quality and test efficiency can be improved. The experimental results show that on fifteen programs, the proposed method has better improvements in terms of indicators such as the coverage rate and average evolutionary generation. Specifically, in terms of the coverage rate, the proposed method demonstrates an improvement of at least 7% and up to 49% compared to three types of baseline methods, and shows the enhancement of approximately 10% to a maximum of 25% compared to five competitive methods. The proposed method provides guidance for the research on multi-path testing that combines machine learning.
Available online: January 14, 2026
, DOI: 10.13328/j.cnki.jos.007558
Abstract:
K-clique enumeration is an important problem in subgraph matching, and the bitmap algorithm has been proven to be an effective method for solving the K-clique enumeration problem. Currently, state-of-the-art K-clique enumeration algorithms are accelerated by GPU. Previous studies have not investigated the impact of sparsity in real-world graph data on bitmap-based K-clique enumeration algorithms. Instead, static parallelization methods and bitmap construction strategies are commonly used on GPU, which result in low computational efficiency. This study proposes a thread-parallel load-balancing scheduling algorithm for bitmap tasks, which resolves the thread divergence problem while achieving high parallelism in the bitmap algorithm. Furthermore, it introduces a dynamic bitmap construction algorithm, enabling bitmaps to be constructed and activated at appropriate times for efficient execution of the bitmap algorithm. A GPU-friendly K-clique enumeration system, KCMiner, is implemented, which adaptively selects optimization strategies for K-clique enumeration tasks. Experimental results on GPU platforms show that the proposed method achieves up to 7.36 times speedup over the baseline K-clique enumeration algorithm and up to 30.2 times speedup over the baseline subgraph matching system.
K-clique enumeration is an important problem in subgraph matching, and the bitmap algorithm has been proven to be an effective method for solving the K-clique enumeration problem. Currently, state-of-the-art K-clique enumeration algorithms are accelerated by GPU. Previous studies have not investigated the impact of sparsity in real-world graph data on bitmap-based K-clique enumeration algorithms. Instead, static parallelization methods and bitmap construction strategies are commonly used on GPU, which result in low computational efficiency. This study proposes a thread-parallel load-balancing scheduling algorithm for bitmap tasks, which resolves the thread divergence problem while achieving high parallelism in the bitmap algorithm. Furthermore, it introduces a dynamic bitmap construction algorithm, enabling bitmaps to be constructed and activated at appropriate times for efficient execution of the bitmap algorithm. A GPU-friendly K-clique enumeration system, KCMiner, is implemented, which adaptively selects optimization strategies for K-clique enumeration tasks. Experimental results on GPU platforms show that the proposed method achieves up to 7.36 times speedup over the baseline K-clique enumeration algorithm and up to 30.2 times speedup over the baseline subgraph matching system.
Available online: January 14, 2026
, DOI: 10.13328/j.cnki.jos.007536
Abstract:
Scade is a well-known commercial tool widely used in the development of safety-critical embedded control software, whose modeling language is a synchronous language extended from Lustre, a synchronous data-flow language. Correct compilation of synchronous languages, including Lustre, has attracted much attention in recent years, and has been addressed in many studies through formal verification. To build a formally verified compiler for such a language, it is a common practice to compile the source program into a C-like program first, and then to compile it into low-level machine-dependent code using a formally verified backend compiler such as the CompCert compiler, where the correct compilation of temporal operators is crucial. In this study, the formally verified compilation of Scade-like temporal operators is introduced, which is used in a formally verified compiler projects, where a Lustre-extended synchronous language is translated into the front-end intermediate language Clight in the CompCert compiler. The compilation and formal verification of temporal operators are divided into two key stages, which are implemented in the interactive proof assistant Coq.
Scade is a well-known commercial tool widely used in the development of safety-critical embedded control software, whose modeling language is a synchronous language extended from Lustre, a synchronous data-flow language. Correct compilation of synchronous languages, including Lustre, has attracted much attention in recent years, and has been addressed in many studies through formal verification. To build a formally verified compiler for such a language, it is a common practice to compile the source program into a C-like program first, and then to compile it into low-level machine-dependent code using a formally verified backend compiler such as the CompCert compiler, where the correct compilation of temporal operators is crucial. In this study, the formally verified compilation of Scade-like temporal operators is introduced, which is used in a formally verified compiler projects, where a Lustre-extended synchronous language is translated into the front-end intermediate language Clight in the CompCert compiler. The compilation and formal verification of temporal operators are divided into two key stages, which are implemented in the interactive proof assistant Coq.
Available online: January 07, 2026
, DOI: 10.13328/j.cnki.jos.007554
Abstract:
The popularization of GPS mobile devices and 5G Internet technology has led to the rapid growth of trajectory data. How to efficiently store, manage, and analyze massive trajectory data has become a hot research issue in the current environment. The traditional single-node trajectory index is limited by memory capacity, disk I/O speed, and other factors, and is no longer capable of managing large-scale trajectory data. Spark, as a distributed framework based on in-memory computing, has natural advantages in processing massive data. Therefore, this study proposes a distributed trajectory data indexing and query scheme based on the Spark platform. To improve the data storage capacity of a single node in a distributed cluster and the efficiency of trajectory queries, a trajectory encoding technique, Z-order trajectory encoding (ZTE), is proposed. This technique encodes the minimum adjacent subspaces covered by the trajectory minimum bounding rectangle (MBR), which can represent trajectories of different granularities and their movement directions, and is used to determine the relationship between a trajectory and the query space. Based on this technique, this study further organizes the ZTE codes of trajectories into a partial-order structure and designs a subspace partial-order branch (SPB). Combined with the hash mapping table IDMap, a local index is constructed. This index avoids the inefficiency caused by the dead space formed by the overlapping of minimum bounding rectangles in R-tree-like indexes and enables fast pruning. To support efficient retrieval of massive trajectory data, the study designs a distributed trajectory index named SPBSpark based on the SPB-branch local index. SPBSpark mainly consists of three components: data partition, local index, and global index. The proposed index effectively supports three types of queries: spatiotemporal range query, k-nearest neighbor query, and moving object trajectory query. Finally, the study selects the distributed trajectory indexes TrajSpark and LocationSpark, which are also based on the Spark framework, as comparison systems. Through comparative simulation experiments, the spatial utilization of the SPBSpark index is improved by about 15% compared with LocationSpark. In terms of query performance, SPBSpark achieves a 2–3 times performance improvement compared with TrajSpark and LocationSpark.
The popularization of GPS mobile devices and 5G Internet technology has led to the rapid growth of trajectory data. How to efficiently store, manage, and analyze massive trajectory data has become a hot research issue in the current environment. The traditional single-node trajectory index is limited by memory capacity, disk I/O speed, and other factors, and is no longer capable of managing large-scale trajectory data. Spark, as a distributed framework based on in-memory computing, has natural advantages in processing massive data. Therefore, this study proposes a distributed trajectory data indexing and query scheme based on the Spark platform. To improve the data storage capacity of a single node in a distributed cluster and the efficiency of trajectory queries, a trajectory encoding technique, Z-order trajectory encoding (ZTE), is proposed. This technique encodes the minimum adjacent subspaces covered by the trajectory minimum bounding rectangle (MBR), which can represent trajectories of different granularities and their movement directions, and is used to determine the relationship between a trajectory and the query space. Based on this technique, this study further organizes the ZTE codes of trajectories into a partial-order structure and designs a subspace partial-order branch (SPB). Combined with the hash mapping table IDMap, a local index is constructed. This index avoids the inefficiency caused by the dead space formed by the overlapping of minimum bounding rectangles in R-tree-like indexes and enables fast pruning. To support efficient retrieval of massive trajectory data, the study designs a distributed trajectory index named SPBSpark based on the SPB-branch local index. SPBSpark mainly consists of three components: data partition, local index, and global index. The proposed index effectively supports three types of queries: spatiotemporal range query, k-nearest neighbor query, and moving object trajectory query. Finally, the study selects the distributed trajectory indexes TrajSpark and LocationSpark, which are also based on the Spark framework, as comparison systems. Through comparative simulation experiments, the spatial utilization of the SPBSpark index is improved by about 15% compared with LocationSpark. In terms of query performance, SPBSpark achieves a 2–3 times performance improvement compared with TrajSpark and LocationSpark.
Available online: January 07, 2026
, DOI: 10.13328/j.cnki.jos.007555
Abstract:
Crowd intelligence is a crucial component of the next generation of artificial intelligence. Researching how to stimulate and converge the innovative power of “people” in open-source communities can significantly enhance development efficiency. Community detection, as a technical approach to studying the relationships among developers in open-source projects, plays a vital role in exploring and understanding social networks. However, current research has predominantly focused on large-scale social networks such as Facebook, while systematic studies on community detection in project-level open source software developer social networks (OSS-DSN) remain limited. This study first collects real-world data and analyzes the features of OSS-DSN. Then, it benchmarks several overlapping and non-overlapping community detection algorithms on these real datasets, comparing algorithm performance across multiple metrics and dimensions. Finally, based on synthetic OSS-DSN, it generates networks efficiently and performs algorithm evaluations using ground-truth data for comparative analysis. Differences in characteristics between small- and medium-scale social networks and large-scale networks are identified, and the influence of these differences on community detection metrics and algorithm performance is explored. The study provides a new benchmark and offers important insights into communication and collaboration in open-source software communities.
Crowd intelligence is a crucial component of the next generation of artificial intelligence. Researching how to stimulate and converge the innovative power of “people” in open-source communities can significantly enhance development efficiency. Community detection, as a technical approach to studying the relationships among developers in open-source projects, plays a vital role in exploring and understanding social networks. However, current research has predominantly focused on large-scale social networks such as Facebook, while systematic studies on community detection in project-level open source software developer social networks (OSS-DSN) remain limited. This study first collects real-world data and analyzes the features of OSS-DSN. Then, it benchmarks several overlapping and non-overlapping community detection algorithms on these real datasets, comparing algorithm performance across multiple metrics and dimensions. Finally, based on synthetic OSS-DSN, it generates networks efficiently and performs algorithm evaluations using ground-truth data for comparative analysis. Differences in characteristics between small- and medium-scale social networks and large-scale networks are identified, and the influence of these differences on community detection metrics and algorithm performance is explored. The study provides a new benchmark and offers important insights into communication and collaboration in open-source software communities.
Available online: January 07, 2026
, DOI: 10.13328/j.cnki.jos.007556
Abstract:
Many code files become oversized and take on excessive responsibilities as software evolves, which severely affects software maintainability and comprehensibility. Developers often need to refactor such files by decomposing a large code file into several smaller ones. Existing studies mainly focus on class file decomposition and are not fully applicable to decomposing complex header files. This is because header file decomposition faces unique challenges. It needs to consider the build dependencies of the entire software project to reduce compilation cost and ensure that the decomposed files are free of cyclic dependencies. To address these challenges, this study proposes an automated approach for decomposing and refactoring complex header files, HeaderSplit. It first constructs a code element graph that captures multiple types of code relationships, including co-usage relationships that reflect project build dependencies. Then, a node coarsening process and a multi-view graph clustering algorithm are applied to identify clusters of closely related code elements. A heuristic algorithm is further introduced to eliminate cyclic dependencies in the clustering results. After the decomposition plan is confirmed, HeaderSplit automatically performs the refactoring, generating new sub-header files and updating the include statements in all code files that directly or indirectly include the original header file. HeaderSplit is evaluated on both synthetic and real complex header files. The results are as follows. 1) HeaderSplit improves accuracy by 11.5% compared with existing methods and demonstrates higher cross-project stability. 2) The decomposed sub-files have higher Modularity and no cyclic dependencies, indicating better architectural design. 3) Using HeaderSplit to decompose complex header files can reduce recompilation costs in their evolution history by15%–60%. 4) HeaderSplit efficiently performs automated refactoring, completing the decomposition and refactoring of header files in large-scale software projects with millions of lines of code within five minutes, showing high practical value.
Many code files become oversized and take on excessive responsibilities as software evolves, which severely affects software maintainability and comprehensibility. Developers often need to refactor such files by decomposing a large code file into several smaller ones. Existing studies mainly focus on class file decomposition and are not fully applicable to decomposing complex header files. This is because header file decomposition faces unique challenges. It needs to consider the build dependencies of the entire software project to reduce compilation cost and ensure that the decomposed files are free of cyclic dependencies. To address these challenges, this study proposes an automated approach for decomposing and refactoring complex header files, HeaderSplit. It first constructs a code element graph that captures multiple types of code relationships, including co-usage relationships that reflect project build dependencies. Then, a node coarsening process and a multi-view graph clustering algorithm are applied to identify clusters of closely related code elements. A heuristic algorithm is further introduced to eliminate cyclic dependencies in the clustering results. After the decomposition plan is confirmed, HeaderSplit automatically performs the refactoring, generating new sub-header files and updating the include statements in all code files that directly or indirectly include the original header file. HeaderSplit is evaluated on both synthetic and real complex header files. The results are as follows. 1) HeaderSplit improves accuracy by 11.5% compared with existing methods and demonstrates higher cross-project stability. 2) The decomposed sub-files have higher Modularity and no cyclic dependencies, indicating better architectural design. 3) Using HeaderSplit to decompose complex header files can reduce recompilation costs in their evolution history by15%–60%. 4) HeaderSplit efficiently performs automated refactoring, completing the decomposition and refactoring of header files in large-scale software projects with millions of lines of code within five minutes, showing high practical value.
Available online: December 24, 2025
, DOI: 10.13328/j.cnki.jos.007557
Abstract:
In software engineering, eliciting non-functional requirements (NFR) remains a critical yet often overlooked task in requirements engineering practice. Traditional NFR elicitation methods predominantly rely on the experience and manual analysis of requirements engineers, leading to inefficiency, omissions, and inconsistencies. Recent breakthroughs in large language models (LLM) in natural language processing have provided new technological means for the automated NFR elicitation. However, directly employing LLM for NFR generation often faces challenges such as hallucination and insufficient domain expertise. To address these issues, this study proposes an automated NFR elicitation method based on LLM to achieve high-quality NFR generation. A structured and correlated dataset comprising 3856 functional requirements and 5723 NFR is constructed, establishing 22647 FR-NFR association pairs. The proposed method integrates retrieval-augmented generation (RAG) technology through three core modules: a semantic case retrieval module based on the maximum marginal relevance algorithm, a prompt engineering module designed for NFR generation, and an optimized LLM generation module. Through professional evaluation by software engineering experts and automatic metrics including BLEU and ROUGE, experimental results demonstrate that the proposed method outperforms existing approaches in terms of completeness, accuracy, and testability of requirements.
In software engineering, eliciting non-functional requirements (NFR) remains a critical yet often overlooked task in requirements engineering practice. Traditional NFR elicitation methods predominantly rely on the experience and manual analysis of requirements engineers, leading to inefficiency, omissions, and inconsistencies. Recent breakthroughs in large language models (LLM) in natural language processing have provided new technological means for the automated NFR elicitation. However, directly employing LLM for NFR generation often faces challenges such as hallucination and insufficient domain expertise. To address these issues, this study proposes an automated NFR elicitation method based on LLM to achieve high-quality NFR generation. A structured and correlated dataset comprising 3856 functional requirements and 5723 NFR is constructed, establishing 22647 FR-NFR association pairs. The proposed method integrates retrieval-augmented generation (RAG) technology through three core modules: a semantic case retrieval module based on the maximum marginal relevance algorithm, a prompt engineering module designed for NFR generation, and an optimized LLM generation module. Through professional evaluation by software engineering experts and automatic metrics including BLEU and ROUGE, experimental results demonstrate that the proposed method outperforms existing approaches in terms of completeness, accuracy, and testability of requirements.
Available online: November 26, 2025
, DOI: 10.13328/j.cnki.jos.007505
Abstract:
HTAP databases are capable of simultaneously supporting OLTP and OLAP workloads within a set of systems. The workload identification is a critical entry point for routing distribution in query execution. The only way to reasonably optimize the queries and allocate resources is to accurately identify whether a query belongs to OLTP or OLAP. Therefore, accurate identification of workload types is a key factor in the performance of HTAP databases. However, existing workload identification methods are mainly based on rules and cost-based measures in SQL statements, as well as machine learning approaches to differentiate workloads. These methods do not consider the inherent characteristics of query statements and utilize structural information in execution plans, resulting in low workload identification accuracy. To improve workload identification accuracy, this study proposes an intelligent method for identifying OLTP and OLAP workloads. This method extracts and encodes features from SQL statements and execution plans, builds the SQL statement encoder based on BERT, and combines the convolutional neural networks and attention mechanisms to construct the encoder of execution plans, with two types of features integrated to build a classifier. The model enables intelligent identification of workloads in HTAP hybrid workloads. Experimental verification shows that the proposed model can accurately identify OLTP and OLAP workloads with high identification accuracy. Additionally, the robustness of the model has been validated across multiple datasets, and the model is integrated into the TiDB database to verify its performance improvement on the database.
HTAP databases are capable of simultaneously supporting OLTP and OLAP workloads within a set of systems. The workload identification is a critical entry point for routing distribution in query execution. The only way to reasonably optimize the queries and allocate resources is to accurately identify whether a query belongs to OLTP or OLAP. Therefore, accurate identification of workload types is a key factor in the performance of HTAP databases. However, existing workload identification methods are mainly based on rules and cost-based measures in SQL statements, as well as machine learning approaches to differentiate workloads. These methods do not consider the inherent characteristics of query statements and utilize structural information in execution plans, resulting in low workload identification accuracy. To improve workload identification accuracy, this study proposes an intelligent method for identifying OLTP and OLAP workloads. This method extracts and encodes features from SQL statements and execution plans, builds the SQL statement encoder based on BERT, and combines the convolutional neural networks and attention mechanisms to construct the encoder of execution plans, with two types of features integrated to build a classifier. The model enables intelligent identification of workloads in HTAP hybrid workloads. Experimental verification shows that the proposed model can accurately identify OLTP and OLAP workloads with high identification accuracy. Additionally, the robustness of the model has been validated across multiple datasets, and the model is integrated into the TiDB database to verify its performance improvement on the database.
Available online: October 29, 2025
, DOI: 10.13328/j.cnki.jos.007477
Abstract:
With the rapid development of the HarmonyOS ecosystem, security issues related to HarmonyOS applications have gradually become a key research focus. In the Android domain, various mature static analysis frameworks have been widely applied to security detection tasks. However, static analysis frameworks for HarmonyOS applications are still in the early stages of development. The OpenHarmony community is currently working on static analysis based on the source code of HarmonyOS applications using ArkTS. However, in practical security detection tasks, obtaining application source code is often difficult, which limits the applicability of this approach. To address this challenge, this study proposes a static analysis framework for HarmonyOS applications based on the Ark intermediate representation (Panda IR). This framework provides basic information interfaces for Panda IR, designs a field-sensitive pointer analysis algorithm tailored to ArkTS syntax features, and implements extended analysis interfaces that interact with pointer analysis. Specifically, 318 instructions in Panda IR are semantically categorized and processed, and a customized pointer flow graph design is further developed. To support ArkTS syntax features, new propagation rules for pointer sets are introduced, and the semantics of special calls are accurately modeled. In addition, based on the pointer analysis results, inter-procedural data dependencies are optimized, and alias analysis capabilities are provided.The experimental evaluation of HarmonyFlow covers three aspects: ArkTS syntax feature coverage, pointer analysis accuracy, and pointer analysis speed. Experimental results show that HarmonyFlow can correctly handle key ArkTS syntax features. The precision and recall rates for call-edge identification in 9 open-source HarmonyOS applications are 98.33% and 92.22%, respectively, with an average runtime of 96 s for 35 real-world HarmonyOS applications.
With the rapid development of the HarmonyOS ecosystem, security issues related to HarmonyOS applications have gradually become a key research focus. In the Android domain, various mature static analysis frameworks have been widely applied to security detection tasks. However, static analysis frameworks for HarmonyOS applications are still in the early stages of development. The OpenHarmony community is currently working on static analysis based on the source code of HarmonyOS applications using ArkTS. However, in practical security detection tasks, obtaining application source code is often difficult, which limits the applicability of this approach. To address this challenge, this study proposes a static analysis framework for HarmonyOS applications based on the Ark intermediate representation (Panda IR). This framework provides basic information interfaces for Panda IR, designs a field-sensitive pointer analysis algorithm tailored to ArkTS syntax features, and implements extended analysis interfaces that interact with pointer analysis. Specifically, 318 instructions in Panda IR are semantically categorized and processed, and a customized pointer flow graph design is further developed. To support ArkTS syntax features, new propagation rules for pointer sets are introduced, and the semantics of special calls are accurately modeled. In addition, based on the pointer analysis results, inter-procedural data dependencies are optimized, and alias analysis capabilities are provided.The experimental evaluation of HarmonyFlow covers three aspects: ArkTS syntax feature coverage, pointer analysis accuracy, and pointer analysis speed. Experimental results show that HarmonyFlow can correctly handle key ArkTS syntax features. The precision and recall rates for call-edge identification in 9 open-source HarmonyOS applications are 98.33% and 92.22%, respectively, with an average runtime of 96 s for 35 real-world HarmonyOS applications.
Available online: September 28, 2025
, DOI: 10.13328/j.cnki.jos.007476
Abstract:
The black-box vulnerability scanner is an essential tool for Web application vulnerability detection, capable of identifying potential security threats effectively before a Web application is launched, thus enhancing the overall security of the application. However, most current black-box scanners primarily collect the attack surface through user operation simulation and regular expression matching. The simulation of user operations is vulnerable to interception by input validation mechanisms and struggles with handling complex event operations, while regular expression matching is ineffective in processing dynamic content. As a result, the scanner cannot effectively address hidden attack surfaces within JavaScript code or dynamically generated attack surfaces, leading to suboptimal vulnerability detection in some Web applications. To resolve these issues, this study proposes a JavaScript Exposure Scanner (JSEScan), a vulnerability scanner enhancement framework based on JavaScript code analysis. The framework integrates static and dynamic code analysis techniques, bypassing form validation and event-triggering restrictions. By extracting attack surface features from JavaScript code, JSEScan identifies attack surfaces and synchronizes them across multiple scanners, enhancing their vulnerability detection capabilities. The experimental results demonstrate that JSEScan increases coverage by 81.02% to 242.15% compared to using a single scanner and uncovers an additional 239 security vulnerabilities when compared to multiple scanners working concurrently, showing superior attack surface collection and vulnerability detection capabilities.
The black-box vulnerability scanner is an essential tool for Web application vulnerability detection, capable of identifying potential security threats effectively before a Web application is launched, thus enhancing the overall security of the application. However, most current black-box scanners primarily collect the attack surface through user operation simulation and regular expression matching. The simulation of user operations is vulnerable to interception by input validation mechanisms and struggles with handling complex event operations, while regular expression matching is ineffective in processing dynamic content. As a result, the scanner cannot effectively address hidden attack surfaces within JavaScript code or dynamically generated attack surfaces, leading to suboptimal vulnerability detection in some Web applications. To resolve these issues, this study proposes a JavaScript Exposure Scanner (JSEScan), a vulnerability scanner enhancement framework based on JavaScript code analysis. The framework integrates static and dynamic code analysis techniques, bypassing form validation and event-triggering restrictions. By extracting attack surface features from JavaScript code, JSEScan identifies attack surfaces and synchronizes them across multiple scanners, enhancing their vulnerability detection capabilities. The experimental results demonstrate that JSEScan increases coverage by 81.02% to 242.15% compared to using a single scanner and uncovers an additional 239 security vulnerabilities when compared to multiple scanners working concurrently, showing superior attack surface collection and vulnerability detection capabilities.
Available online: October 18, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表Proceedings of the 11th Joint Meeting of the European Software Engineering Conference and the ACM SigSoft Symposium on The Foundations of Software Engineering (ESEC/FSE)，ACM，2017年9月，315-325页. 原文链接如下：https://doi.org/10.1145/3106237.3106242， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表Proceedings of the 11th Joint Meeting of the European Software Engineering Conference and the ACM SigSoft Symposium on The Foundations of Software Engineering (ESEC/FSE)，ACM，2017年9月，315-325页. 原文链接如下：https://doi.org/10.1145/3106237.3106242， 读者如需引用该文请标引原文出处。
Available online: October 18, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表Proceedings of the 11th Joint Meeting of the European Software Engineering Conference and the ACM SigSoft Symposium on The Foundations of Software Engineering (ESEC/FSE)，ACM，2017年9月，303-314页. 原文链接如下：https://doi.org/10.1145/3106237.3106239， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表Proceedings of the 11th Joint Meeting of the European Software Engineering Conference and the ACM SigSoft Symposium on The Foundations of Software Engineering (ESEC/FSE)，ACM，2017年9月，303-314页. 原文链接如下：https://doi.org/10.1145/3106237.3106239， 读者如需引用该文请标引原文出处。
Available online: September 11, 2017
Abstract:
GitHub, a popular social-software-development platform, has fostered a variety of software ecosystems where projects depend on one another and practitioners interact with each other. Projects within an ecosystem often have complex inter-dependencies that impose new challenges in bug reporting and fixing. In this paper, we conduct an empirical study on cross-project correlated bugs, i.e., causally related bugs reported to different projects, focusing on two aspects: 1) how developers track the root causes across projects; and 2) how the downstream developers coordinate to deal with upstream bugs. Through manual inspection of bug reports collected from the scientific Python ecosystem and an online survey with developers, this study reveals the common practices of developers and the various factors in fixing cross-project bugs. These findings provide implications for future software bug analysis in the scope of ecosystem, as well as shed light on the requirements of issue trackers for such bugs.
GitHub, a popular social-software-development platform, has fostered a variety of software ecosystems where projects depend on one another and practitioners interact with each other. Projects within an ecosystem often have complex inter-dependencies that impose new challenges in bug reporting and fixing. In this paper, we conduct an empirical study on cross-project correlated bugs, i.e., causally related bugs reported to different projects, focusing on two aspects: 1) how developers track the root causes across projects; and 2) how the downstream developers coordinate to deal with upstream bugs. Through manual inspection of bug reports collected from the scientific Python ecosystem and an online survey with developers, this study reveals the common practices of developers and the various factors in fixing cross-project bugs. These findings provide implications for future software bug analysis in the scope of ecosystem, as well as shed light on the requirements of issue trackers for such bugs.
Available online: June 21, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在IEEE Transactions on Software Engineering 2017 已录用待发表. 原文链接如下：http://ieeexplore.ieee.org/document/7792694， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在IEEE Transactions on Software Engineering 2017 已录用待发表. 原文链接如下：http://ieeexplore.ieee.org/document/7792694， 读者如需引用该文请标引原文出处。
Available online: June 13, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 39th International Conference on Software Engineering, Pages 27-37, Buenos Aires, Argentina — May 20 - 28, 2017, IEEE Press Piscataway, NJ, USA ?2017, ISBN: 978-1-5386-3868-2 原文链接如下：http://dl.acm.org/citation.cfm?id=3097373， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 39th International Conference on Software Engineering, Pages 27-37, Buenos Aires, Argentina — May 20 - 28, 2017, IEEE Press Piscataway, NJ, USA ?2017, ISBN: 978-1-5386-3868-2 原文链接如下：http://dl.acm.org/citation.cfm?id=3097373， 读者如需引用该文请标引原文出处。
Available online: January 25, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering (FSE 2016). ACM, New York, NY, USA, 871-882. DOI: https://doi.org/10.1145/2950290.2950364 原文链接如下：http://dl.acm.org/citation.cfm?id=2950364， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering (FSE 2016). ACM, New York, NY, USA, 871-882. DOI: https://doi.org/10.1145/2950290.2950364 原文链接如下：http://dl.acm.org/citation.cfm?id=2950364， 读者如需引用该文请标引原文出处。
Available online: January 18, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering, Pages 133—143, Seattle WA, USA, November 2016. 原文链接如下：http://dl.acm.org/citation.cfm?id=2950327， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering, Pages 133—143, Seattle WA, USA, November 2016. 原文链接如下：http://dl.acm.org/citation.cfm?id=2950327， 读者如需引用该文请标引原文出处。
Available online: January 04, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering (FSE'16), 810 – 821, November 13 - 18, 2016. 原文链接如下：https://doi.org/10.1145/2950290.2950310， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering (FSE'16), 810 – 821, November 13 - 18, 2016. 原文链接如下：https://doi.org/10.1145/2950290.2950310， 读者如需引用该文请标引原文出处。
Available online: January 04, 2017
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在FSE 2016, 原文链接如下：http://dl.acm.org/citation.cfm?doid=2950290.2950313, 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在FSE 2016, 原文链接如下：http://dl.acm.org/citation.cfm?doid=2950290.2950313, 读者如需引用该文请标引原文出处。
Available online: December 12, 2016
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在FSE'16会议上Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering， 原文链接如下：http://dl.acm.org/citation.cfm?id=2950340， 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在FSE'16会议上Proceedings of the 2016 24th ACM SIGSOFT International Symposium on Foundations of Software Engineering， 原文链接如下：http://dl.acm.org/citation.cfm?id=2950340， 读者如需引用该文请标引原文出处。
Available online: September 30, 2016
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在ASE2016会议上http://ase2016.org/ 原文链接如下：http://dl.acm.org/citation.cfm?id=2970366 读者如需引用该文请标引原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 文章发表在ASE2016会议上http://ase2016.org/ 原文链接如下：http://dl.acm.org/citation.cfm?id=2970366 读者如需引用该文请标引原文出处。
Available online: September 09, 2016
Abstract:
文章由CCF软件工程专业委员会白颖教授推荐。 俊杰的文章发表在ASE2016会议上，http://ase2016.org/。 原文链接如下：http://dl.acm.org/citation.cfm?doid=2970276.2970300 请读者标引时请引注原文出处。
文章由CCF软件工程专业委员会白颖教授推荐。 俊杰的文章发表在ASE2016会议上，http://ase2016.org/。 原文链接如下：http://dl.acm.org/citation.cfm?doid=2970276.2970300 请读者标引时请引注原文出处。
Available online: September 07, 2016
Abstract:
CCF 软件工程专业委员会白晓颖教授（清华大学）推荐。 原文发表在 ASE 2016 Proceedings of the 31st IEEE/ACM International Conference on Automated Software Engineering。 全文链接：http://dx.doi.org/10.1145/2970276.2970307。 重要提示：读者如引用该文时请标注原文出处。
CCF 软件工程专业委员会白晓颖教授（清华大学）推荐。 原文发表在 ASE 2016 Proceedings of the 31st IEEE/ACM International Conference on Automated Software Engineering。 全文链接：http://dx.doi.org/10.1145/2970276.2970307。 重要提示：读者如引用该文时请标注原文出处。
Available online: August 29, 2016
Abstract:
CCF软件工程专业委员会白晓颖教授（清华大学）推荐。 该论文发表在ACM Transactions on Software Engineering and Methodology （TOSEM, Vol. 25, No. 2, Article 13, May 2016），被ICSE 2016主会邀请为“Journal first”报告， 全文参见http://dl.acm.org/citation.cfm?id=2876443。 论文作者包括北京大学的周明辉,马秀娟,张路和梅宏，以及田纳西大学的Audris Mockus。 重要提示：读者如引用该文时请标注原文出处。
CCF软件工程专业委员会白晓颖教授（清华大学）推荐。 该论文发表在ACM Transactions on Software Engineering and Methodology （TOSEM, Vol. 25, No. 2, Article 13, May 2016），被ICSE 2016主会邀请为“Journal first”报告， 全文参见http://dl.acm.org/citation.cfm?id=2876443。 论文作者包括北京大学的周明辉,马秀娟,张路和梅宏，以及田纳西大学的Audris Mockus。 重要提示：读者如引用该文时请标注原文出处。
-
全文下载排行(总排行，年度排行，各期排行)
摘要点击排行(总排行，年度排行，各期排行)
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2015,26(6):1356-1372
, DOI: 10.13328/j.cnki.jos.004831
Abstract:
Social recommender systems have recently become one of the hottest topics in the domain of recommender systems. The main task of social recommender system is to alleviate data sparsity and cold-start problems, and improve its performance utilizing users' social attributes. This paper presents an overview of the field of social recommender systems, including trust inference algorithms, key techniques and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
Social recommender systems have recently become one of the hottest topics in the domain of recommender systems. The main task of social recommender system is to alleviate data sparsity and cold-start problems, and improve its performance utilizing users' social attributes. This paper presents an overview of the field of social recommender systems, including trust inference algorithms, key techniques and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
2019,30(1):22-32
, DOI: 10.13328/j.cnki.jos.005648
Abstract:
This paper presents several new insights into system software, which is one of the basic concepts in computing discipline, from three perspectives of essential features, characteristics of the times, and the future development trend. The first insight is that system software stems theoretically and technically from universal Turing machine and the idea of stored-program, with an essential feature of "manipulating the execution of a computing system". There are two typical manipulation modes:encoding and then loading, executing and controlling. The second insight is that software system is a kind of software, in the Internet age, providing substantial online services continuously, which lay the foundation for the newly emerged "software-as-a-service" paradigm. The final insight is about its development trend:system software will evolve online continuously. Driven by innovations of computing systems, integration of cyber and physical spaces, and intelligence technologies, system software will become the core of future software ecology.
This paper presents several new insights into system software, which is one of the basic concepts in computing discipline, from three perspectives of essential features, characteristics of the times, and the future development trend. The first insight is that system software stems theoretically and technically from universal Turing machine and the idea of stored-program, with an essential feature of "manipulating the execution of a computing system". There are two typical manipulation modes:encoding and then loading, executing and controlling. The second insight is that software system is a kind of software, in the Internet age, providing substantial online services continuously, which lay the foundation for the newly emerged "software-as-a-service" paradigm. The final insight is about its development trend:system software will evolve online continuously. Driven by innovations of computing systems, integration of cyber and physical spaces, and intelligence technologies, system software will become the core of future software ecology.
2016,27(6):1328-1348
, DOI: 10.13328/j.cnki.jos.005004
Abstract:
With the rapid development of cloud computing technology, its security issues have become more and more obvious and received much attention in both industry and academia. High security risk is widespread in traditional cloud architecture. Hacking into a virtual machine destroys the availability of cloud services or resources. Un-Trusted cloud storage makes it more difficult to share or search users' private data. The risk of privacy leakage is caused by various outsourcing computation and application requirements. From the perspective of security and privacy preserving technologies in cloud computing, this paper first introduces related research progress of cloud virtualization security, cloud data security and cloud application security. In addition, it analyzes the characteristics and application scopes of typical schemes, and compares their different effectiveness on the security defense and privacy preserving. Finally, the paper discusses current limitations and possible directions for future research.
With the rapid development of cloud computing technology, its security issues have become more and more obvious and received much attention in both industry and academia. High security risk is widespread in traditional cloud architecture. Hacking into a virtual machine destroys the availability of cloud services or resources. Un-Trusted cloud storage makes it more difficult to share or search users' private data. The risk of privacy leakage is caused by various outsourcing computation and application requirements. From the perspective of security and privacy preserving technologies in cloud computing, this paper first introduces related research progress of cloud virtualization security, cloud data security and cloud application security. In addition, it analyzes the characteristics and application scopes of typical schemes, and compares their different effectiveness on the security defense and privacy preserving. Finally, the paper discusses current limitations and possible directions for future research.
2015,26(1):26-39
, DOI: 10.13328/j.cnki.jos.004631
Abstract:
In recent years, transfer learning has provoked vast amount of attention and research. Transfer learning is a new machine learning method that applies the knowledge from related but different domains to target domains. It relaxes the two basic assumptions in traditional machine learning: (1) the training (also referred as source domain) and test data (also referred target domain) follow the independent and identically distributed (i.i.d.) condition; (2) there are enough labeled samples to learn a good classification model, aiming to solve the problems that there are few or even not any labeled data in target domains. This paper surveys the research progress of transfer learning and introduces its own works, especially the ones in building transfer learning models by applying generative model on the concept level. Finally, the paper introduces the applications of transfer learning, such as text classification and collaborative filtering, and further suggests the future research direction of transfer learning.
In recent years, transfer learning has provoked vast amount of attention and research. Transfer learning is a new machine learning method that applies the knowledge from related but different domains to target domains. It relaxes the two basic assumptions in traditional machine learning: (1) the training (also referred as source domain) and test data (also referred target domain) follow the independent and identically distributed (i.i.d.) condition; (2) there are enough labeled samples to learn a good classification model, aiming to solve the problems that there are few or even not any labeled data in target domains. This paper surveys the research progress of transfer learning and introduces its own works, especially the ones in building transfer learning models by applying generative model on the concept level. Finally, the paper introduces the applications of transfer learning, such as text classification and collaborative filtering, and further suggests the future research direction of transfer learning.
2015,26(1):62-81
, DOI: 10.13328/j.cnki.jos.004701
Abstract:
Network abstraction brings about the naissance of software-defined networking. SDN decouples data plane and control plane, and simplifies network management. The paper starts with a discussion on the background in the naissance and developments of SDN, combing its architecture that includes data layer, control layer and application layer. Then their key technologies are elaborated according to the hierarchical architecture of SDN. The characteristics of consistency, availability, and tolerance are especially analyzed. Moreover, latest achievements for profiled scenes are introduced. The future works are summarized in the end.
Network abstraction brings about the naissance of software-defined networking. SDN decouples data plane and control plane, and simplifies network management. The paper starts with a discussion on the background in the naissance and developments of SDN, combing its architecture that includes data layer, control layer and application layer. Then their key technologies are elaborated according to the hierarchical architecture of SDN. The characteristics of consistency, availability, and tolerance are especially analyzed. Moreover, latest achievements for profiled scenes are introduced. The future works are summarized in the end.
2003,14(7):1282-1291
Abstract:
Sensor network, which is made by the convergence of sensor, micro-electro-mechanism system and networks technologies, is a novel technology about acquiring and processing information. In this paper, the architecture of wireless sensor network is briefly introduced. Next, some valuable applications are explained and forecasted. Combining with the existing work, the hot spots including power-aware routing and media access control schemes are discussed and presented in detail. Finally, taking account of application requirements, several future research directions are put forward.
Sensor network, which is made by the convergence of sensor, micro-electro-mechanism system and networks technologies, is a novel technology about acquiring and processing information. In this paper, the architecture of wireless sensor network is briefly introduced. Next, some valuable applications are explained and forecasted. Combining with the existing work, the hot spots including power-aware routing and media access control schemes are discussed and presented in detail. Finally, taking account of application requirements, several future research directions are put forward.
2010,21(3):427-437
Abstract:
Automatic generation of poetry has always been considered a hard nut in natural language generation.This paper reports some pioneering research on a possible generic algorithm and its automatic generation of SONGCI. In light of the characteristics of Chinese ancient poetry, this paper designed the level and oblique tones-based coding method, the syntactic and semantic weighted function of fitness, the elitism and roulette-combined selection operator, and the partially mapped crossover operator and the heuristic mutation operator. As shown by tests, the system constructed on the basis of the computing model designed in this paper is basically capable of generating Chinese SONGCI with some aesthetic merit. This work represents progress in the field of Chinese poetry automatic generation.
Automatic generation of poetry has always been considered a hard nut in natural language generation.This paper reports some pioneering research on a possible generic algorithm and its automatic generation of SONGCI. In light of the characteristics of Chinese ancient poetry, this paper designed the level and oblique tones-based coding method, the syntactic and semantic weighted function of fitness, the elitism and roulette-combined selection operator, and the partially mapped crossover operator and the heuristic mutation operator. As shown by tests, the system constructed on the basis of the computing model designed in this paper is basically capable of generating Chinese SONGCI with some aesthetic merit. This work represents progress in the field of Chinese poetry automatic generation.
2016,27(3):691-713
, DOI: 10.13328/j.cnki.jos.004948
Abstract:
Learning to rank(L2R) techniques try to solve sorting problems using machine learning methods, and have been well studied and widely used in various fields such as information retrieval, text mining, personalized recommendation, and biomedicine.The main task of L2R based recommendation algorithms is integrating L2R techniques into recommendation algorithms, and studying how to organize a large number of users and features of items, build more suitable user models according to user preferences requirements, and improve the performance and user satisfaction of recommendation algorithms.This paper surveys L2R based recommendation algorithms in recent years, summarizes the problem definition, compares key technologies and analyzes evaluation metrics and their applications.In addition, the paper discusses the future development trend of L2R based recommendation algorithms.
Learning to rank(L2R) techniques try to solve sorting problems using machine learning methods, and have been well studied and widely used in various fields such as information retrieval, text mining, personalized recommendation, and biomedicine.The main task of L2R based recommendation algorithms is integrating L2R techniques into recommendation algorithms, and studying how to organize a large number of users and features of items, build more suitable user models according to user preferences requirements, and improve the performance and user satisfaction of recommendation algorithms.This paper surveys L2R based recommendation algorithms in recent years, summarizes the problem definition, compares key technologies and analyzes evaluation metrics and their applications.In addition, the paper discusses the future development trend of L2R based recommendation algorithms.
2013,24(1):91-108
, DOI: 10.3724/SP.J.1001.2013.04292
Abstract:
Mobile recommender systems have recently become one of the hottest topics in the domain of recommender systems. The main task of mobile recommender systems is to improve the performance and accuracy along with user satisfaction utilizing mobile context, mobile social network and other information. This paper presents an overview of the field of mobile recommender systems including key techniques, evaluation and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
Mobile recommender systems have recently become one of the hottest topics in the domain of recommender systems. The main task of mobile recommender systems is to improve the performance and accuracy along with user satisfaction utilizing mobile context, mobile social network and other information. This paper presents an overview of the field of mobile recommender systems including key techniques, evaluation and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
2011,22(1):71-83
, DOI: 10.3724/SP.J.1001.2011.03958
Abstract:
Cloud Computing is the fundamental change happening in the field of Information Technology. It is a representation of a movement towards the intensive, large scale specialization. On the other hand, it brings about not only convenience and efficiency problems, but also great challenges in the field of data security and privacy protection. Currently, security has been regarded as one of the greatest problems in the development of Cloud Computing. This paper describes the great requirements in Cloud Computing, security key technology, standard and regulation etc., and provides a Cloud Computing security framework. This paper argues that the changes in the above aspects will result in a technical revolution in the field of information security.
Cloud Computing is the fundamental change happening in the field of Information Technology. It is a representation of a movement towards the intensive, large scale specialization. On the other hand, it brings about not only convenience and efficiency problems, but also great challenges in the field of data security and privacy protection. Currently, security has been regarded as one of the greatest problems in the development of Cloud Computing. This paper describes the great requirements in Cloud Computing, security key technology, standard and regulation etc., and provides a Cloud Computing security framework. This paper argues that the changes in the above aspects will result in a technical revolution in the field of information security.
2016,27(1):45-71
, DOI: 10.13328/j.cnki.jos.004914
Abstract:
Android is a modern and most popular software platform for smartphones. According to report, Android accounted for a huge 81% of all smartphones in 2014 and shipped over 1 billion units worldwide for the first time ever. Apple, Microsoft, Blackberry and Firefox trailed a long way behind. At the same time, increased popularity of the Android smartphones has attracted hackers, leading to massive increase of Android malware applications. This paper summarizes and analyzes the latest advances in Android security from multidimensional perspectives, covering Android architecture, design principles, security mechanisms, major security threats, classification and detection of malware, static and dynamic analyses, machine learning approaches, and security extension proposals.
Android is a modern and most popular software platform for smartphones. According to report, Android accounted for a huge 81% of all smartphones in 2014 and shipped over 1 billion units worldwide for the first time ever. Apple, Microsoft, Blackberry and Firefox trailed a long way behind. At the same time, increased popularity of the Android smartphones has attracted hackers, leading to massive increase of Android malware applications. This paper summarizes and analyzes the latest advances in Android security from multidimensional perspectives, covering Android architecture, design principles, security mechanisms, major security threats, classification and detection of malware, static and dynamic analyses, machine learning approaches, and security extension proposals.
2018,29(2):340-362
, DOI: 10.13328/j.cnki.jos.005391
Abstract:
With the increasing of social network, social recommendation becomes hot research topic in recommendation systems. Matrix factorization based (MF-based) recommendation model gradually becomes the key component of social recommendation due to its high expansibility and flexibility. Thus, this paper focuses on MF-based social recommendation methods. Firstly, it reviews the existing social recommendation models according to the model construction strategies. Next, it conducts a series of experiments on real-world datasets to demonstrate the performance of different social recommendation methods from three perspectives including whole-users, cold start-users, and long-tail items. Finally, the paper analyzes the problems of MF-based social recommendation model, and discusses the possible future research directions and development trends in this research area.
With the increasing of social network, social recommendation becomes hot research topic in recommendation systems. Matrix factorization based (MF-based) recommendation model gradually becomes the key component of social recommendation due to its high expansibility and flexibility. Thus, this paper focuses on MF-based social recommendation methods. Firstly, it reviews the existing social recommendation models according to the model construction strategies. Next, it conducts a series of experiments on real-world datasets to demonstrate the performance of different social recommendation methods from three perspectives including whole-users, cold start-users, and long-tail items. Finally, the paper analyzes the problems of MF-based social recommendation model, and discusses the possible future research directions and development trends in this research area.
2008,19(1):48-61
Abstract:
The research actuality and new progress in clustering algorithm in recent years are summarized in this paper. First, the analysis and induction of some representative clustering algorithms have been made from several aspects, such as the ideas of algorithm, key technology, advantage and disadvantage. On the other hand, several typical clustering algorithms and known data sets are selected, simulation experiments are implemented from both sides of accuracy and running efficiency, and clustering condition of one algorithm with different data sets is analyzed by comparing with the same clustering of the data set under different algorithms. Finally, the research hotspot, difficulty, shortage of the data clustering and some pending problems are addressed by the integration of the aforementioned two aspects information. The above work can give a valuable reference for data clustering and data mining.
The research actuality and new progress in clustering algorithm in recent years are summarized in this paper. First, the analysis and induction of some representative clustering algorithms have been made from several aspects, such as the ideas of algorithm, key technology, advantage and disadvantage. On the other hand, several typical clustering algorithms and known data sets are selected, simulation experiments are implemented from both sides of accuracy and running efficiency, and clustering condition of one algorithm with different data sets is analyzed by comparing with the same clustering of the data set under different algorithms. Finally, the research hotspot, difficulty, shortage of the data clustering and some pending problems are addressed by the integration of the aforementioned two aspects information. The above work can give a valuable reference for data clustering and data mining.
2009,20(5):1337-1348
Abstract:
This paper surveys the current technologies adopted in cloud computing as well as the systems in enterprises. Cloud computing can be viewed from two different aspects. One is about the cloud infrastructure which is the building block for the up layer cloud application. The other is of course the cloud application. This paper focuses on the cloud infrastructure including the systems and current research. Some attractive cloud applications are also discussed. Cloud computing infrastructure has three distinct characteristics. First, the infrastructure is built on top of large scale clusters which contain a large number of cheap PC servers. Second, the applications are co-designed with the fundamental infrastructure that the computing resources can be maximally utilized. Third, the reliability of the whole system is achieved by software building on top of redundant hardware instead of mere hardware. All these technologies are for the two important goals for distributed system: high scalability and high availability. Scalability means that the cloud infrastructure can be expanded to very large scale even to thousands of nodes. Availability means that the services are available even when quite a number of nodes fail. From this paper, readers will capture the current status of cloud computing as well as its future trends.
This paper surveys the current technologies adopted in cloud computing as well as the systems in enterprises. Cloud computing can be viewed from two different aspects. One is about the cloud infrastructure which is the building block for the up layer cloud application. The other is of course the cloud application. This paper focuses on the cloud infrastructure including the systems and current research. Some attractive cloud applications are also discussed. Cloud computing infrastructure has three distinct characteristics. First, the infrastructure is built on top of large scale clusters which contain a large number of cheap PC servers. Second, the applications are co-designed with the fundamental infrastructure that the computing resources can be maximally utilized. Third, the reliability of the whole system is achieved by software building on top of redundant hardware instead of mere hardware. All these technologies are for the two important goals for distributed system: high scalability and high availability. Scalability means that the cloud infrastructure can be expanded to very large scale even to thousands of nodes. Availability means that the services are available even when quite a number of nodes fail. From this paper, readers will capture the current status of cloud computing as well as its future trends.
2009,20(2):271-289
Abstract:
Evolutionary multi-objective optimization (EMO), whose main task is to deal with multi-objective optimization problems by evolutionary computation, has become a hot topic in evolutionary computation community. After summarizing the EMO algorithms before 2003 briefly, the recent advances in EMO are discussed in details. The current research directions are concluded. On the one hand, more new evolutionary paradigms have been introduced into EMO community, such as particle swarm optimization, artificial immune systems, and estimation distribution algorithms. On the other hand, in order to deal with many-objective optimization problems, many new dominance schemes different from traditional Pareto-dominance come forth. Furthermore, the essential characteristics of multi-objective optimization problems are deeply investigated. This paper also gives experimental comparison of several representative algorithms. Finally, several viewpoints for the future research of EMO are proposed.
Evolutionary multi-objective optimization (EMO), whose main task is to deal with multi-objective optimization problems by evolutionary computation, has become a hot topic in evolutionary computation community. After summarizing the EMO algorithms before 2003 briefly, the recent advances in EMO are discussed in details. The current research directions are concluded. On the one hand, more new evolutionary paradigms have been introduced into EMO community, such as particle swarm optimization, artificial immune systems, and estimation distribution algorithms. On the other hand, in order to deal with many-objective optimization problems, many new dominance schemes different from traditional Pareto-dominance come forth. Furthermore, the essential characteristics of multi-objective optimization problems are deeply investigated. This paper also gives experimental comparison of several representative algorithms. Finally, several viewpoints for the future research of EMO are proposed.
2018,29(9):2681-2696
, DOI: 10.13328/j.cnki.jos.005274
Abstract:
Recommender systems have been successfully adopted as an effective tool to alleviate information overload and assist users to make decisions. Recently, it has been demonstrated that incorporating social relationships into recommender models can enhance recommendation performance. Despite its remarkable progress, a majority of social recommendation models have overlooked the item relations-a key factor that can also significantly influence recommendation performance. In this paper, a approach is first proposed to acquire item relations by measuring correlations among items. Then, a co-regularized recommendation model is put forward to integrate the item relations with social relationships by introducing co-regularization term in the matrix factorization model. Meanwhile, that the co-regularization term is a case of weighted atomic norm is illustrated. Finally, based on the proposed model a recommendation algorithm named CRMF is constructed. CRMF is compared with existing state-of-the-art recommendation algorithms based on the evaluations over four real-world data sets. The experimental results demonstrate that CRMF is able to not only effectively alleviate the user cold-start problem, but also help obtain more accurate rating predictions of various users.
Recommender systems have been successfully adopted as an effective tool to alleviate information overload and assist users to make decisions. Recently, it has been demonstrated that incorporating social relationships into recommender models can enhance recommendation performance. Despite its remarkable progress, a majority of social recommendation models have overlooked the item relations-a key factor that can also significantly influence recommendation performance. In this paper, a approach is first proposed to acquire item relations by measuring correlations among items. Then, a co-regularized recommendation model is put forward to integrate the item relations with social relationships by introducing co-regularization term in the matrix factorization model. Meanwhile, that the co-regularization term is a case of weighted atomic norm is illustrated. Finally, based on the proposed model a recommendation algorithm named CRMF is constructed. CRMF is compared with existing state-of-the-art recommendation algorithms based on the evaluations over four real-world data sets. The experimental results demonstrate that CRMF is able to not only effectively alleviate the user cold-start problem, but also help obtain more accurate rating predictions of various users.
2018,29(3):786-798
, DOI: 10.13328/j.cnki.jos.005437
Abstract:
Graph embedding is a fundamental technique for graph data mining. The real-world graphs not only consist of complex network structures, but also contain diverse vertex information. How to integrate the network structure and vertex information into the graph embedding procedure is a big challenge. To deal with this challenge, a graph embedding method, which is based on deep leaning technique while taking into account the prior knowledge on vertices information, is proposed in this paper. The basic idea of the proposed method is to regard the vertex features as the prior knowledge, and learn the representation vector through optimizing an objective function that simultaneously keeps the similarity of network structure and vertex features. The time complexity of the proposed method is O(|V|), where|V|is the count of vertices in the graph. This indicates the proposed method is suitable for large-scale graph analysis. Experiments on several data sets demonstrate that, compared with the state-of-art baselines, the proposed method is able to achieve favorable and stable results for the task of node classification.
Graph embedding is a fundamental technique for graph data mining. The real-world graphs not only consist of complex network structures, but also contain diverse vertex information. How to integrate the network structure and vertex information into the graph embedding procedure is a big challenge. To deal with this challenge, a graph embedding method, which is based on deep leaning technique while taking into account the prior knowledge on vertices information, is proposed in this paper. The basic idea of the proposed method is to regard the vertex features as the prior knowledge, and learn the representation vector through optimizing an objective function that simultaneously keeps the similarity of network structure and vertex features. The time complexity of the proposed method is O(|V|), where|V|is the count of vertices in the graph. This indicates the proposed method is suitable for large-scale graph analysis. Experiments on several data sets demonstrate that, compared with the state-of-art baselines, the proposed method is able to achieve favorable and stable results for the task of node classification.
2017,28(4):959-992
, DOI: 10.13328/j.cnki.jos.005143
Abstract:
The development of mobile internet and the popularity of mobile terminals produce massive trajectory data of moving objects under the era of big data. Trajectory data has spatio-temporal characteristics and rich information. Trajectory data processing techniques can be used to mine the patterns of human activities and behaviors, the moving patterns of vehicles in the city and the changes of atmospheric environment. However, trajectory data also can be exploited to disclose moving objects' privacy information (e.g., behaviors, hobbies and social relationships). Accordingly, attackers can easily access moving objects' privacy information by digging into their trajectory data such as activities and check-in locations. In another front of research, quantum computation presents an important theoretical direction to mine big data due to its scalable and powerful storage and computing capacity. Applying quantum computing approaches to handle trajectory big data could make some complex problem solvable and achieve higher efficiency. This paper reviews the key technologies of processing trajectory data. First the concept and characteristics of trajectory data is introduced, and the pre-processing methods, including noise filtering and data compression, are summarized. Then, the trajectory indexing and querying techniques, and the current achievements of mining trajectory data, such as pattern mining and trajectory classification, are reviewed. Next, an overview of the basic theories and characteristics of privacy preserving with respect to trajectory data is provided. The supporting techniques of trajectory big data mining, such as processing framework and data visualization, are presented in detail. Some possible ways of applying quantum computation into trajectory data processing, as well as the implementation of some core trajectory mining algorithms by quantum computation are also described. Finally, the challenges of trajectory data processing and promising future research directions are discussed.
The development of mobile internet and the popularity of mobile terminals produce massive trajectory data of moving objects under the era of big data. Trajectory data has spatio-temporal characteristics and rich information. Trajectory data processing techniques can be used to mine the patterns of human activities and behaviors, the moving patterns of vehicles in the city and the changes of atmospheric environment. However, trajectory data also can be exploited to disclose moving objects' privacy information (e.g., behaviors, hobbies and social relationships). Accordingly, attackers can easily access moving objects' privacy information by digging into their trajectory data such as activities and check-in locations. In another front of research, quantum computation presents an important theoretical direction to mine big data due to its scalable and powerful storage and computing capacity. Applying quantum computing approaches to handle trajectory big data could make some complex problem solvable and achieve higher efficiency. This paper reviews the key technologies of processing trajectory data. First the concept and characteristics of trajectory data is introduced, and the pre-processing methods, including noise filtering and data compression, are summarized. Then, the trajectory indexing and querying techniques, and the current achievements of mining trajectory data, such as pattern mining and trajectory classification, are reviewed. Next, an overview of the basic theories and characteristics of privacy preserving with respect to trajectory data is provided. The supporting techniques of trajectory big data mining, such as processing framework and data visualization, are presented in detail. Some possible ways of applying quantum computation into trajectory data processing, as well as the implementation of some core trajectory mining algorithms by quantum computation are also described. Finally, the challenges of trajectory data processing and promising future research directions are discussed.
2018,29(10):3164-3183
, DOI: 10.13328/j.cnki.jos.005288
Abstract:
Group recommender systems have recently become one of the most prevalent topics in recommender systems. As an effective solution to the problem of group recommendation, Group recommender systems have been utilized in news, music, movies, food, and so forth through extending individual recommendation to group recommendation. The existing group recommender systems usually employ aggregating preference strategy or aggregating recommendation strategy, but the effectiveness of both two methods is not well solved yet, and they respectively have their own advantages and disadvantages. Aggregating preference strategy possesses a fairness problem between group members, whereas aggregating recommendation strategy pays less attention to the interaction between group members. This paper proposes an enhanced group recommendation method based on preference aggregation, incorporating simultaneously the advantages of the aforesaid two aggregation methods. Further, the paper demonstrates that group preference and personal preference are similar, which is also considered in the proposed method. Experimental results show that the proposed method outperforms baselines in terms of effectiveness based on Movielens dataset.
Group recommender systems have recently become one of the most prevalent topics in recommender systems. As an effective solution to the problem of group recommendation, Group recommender systems have been utilized in news, music, movies, food, and so forth through extending individual recommendation to group recommendation. The existing group recommender systems usually employ aggregating preference strategy or aggregating recommendation strategy, but the effectiveness of both two methods is not well solved yet, and they respectively have their own advantages and disadvantages. Aggregating preference strategy possesses a fairness problem between group members, whereas aggregating recommendation strategy pays less attention to the interaction between group members. This paper proposes an enhanced group recommendation method based on preference aggregation, incorporating simultaneously the advantages of the aforesaid two aggregation methods. Further, the paper demonstrates that group preference and personal preference are similar, which is also considered in the proposed method. Experimental results show that the proposed method outperforms baselines in terms of effectiveness based on Movielens dataset.
2018,29(10):3134-3149
, DOI: 10.13328/j.cnki.jos.005284
Abstract:
Event-Based social networks (EBSNs) have experienced rapid growth in people's daily life. Hence, event recommendation plays an important role in helping people discover interesting online events and attend offline activities face to face in the real world. However, event recommendation is quite different from traditional recommender systems, and there are several challenges:(1) One user can only attend a scarce number of events, leading to a very sparse user-event matrix; (2) The response data of users is implicit feedback; (3) Events have their life cycles, so outdated events should not be recommended to users; (4) A large number of new events which are created every day need to be recommended to users in time. To cope with these challenges, this article proposes to jointly model heterogeneous social and content information for event recommendation. This approach explores both the online and offline social interactions and fuses the content of events to model their joint effect on users' decision-making for events. Extensive experiments are conducted to evaluate the performance of the proposed model on Meetup dataset. The experimental results demonstrate that the proposed model outperforms state-of-the-art methods.
Event-Based social networks (EBSNs) have experienced rapid growth in people's daily life. Hence, event recommendation plays an important role in helping people discover interesting online events and attend offline activities face to face in the real world. However, event recommendation is quite different from traditional recommender systems, and there are several challenges:(1) One user can only attend a scarce number of events, leading to a very sparse user-event matrix; (2) The response data of users is implicit feedback; (3) Events have their life cycles, so outdated events should not be recommended to users; (4) A large number of new events which are created every day need to be recommended to users in time. To cope with these challenges, this article proposes to jointly model heterogeneous social and content information for event recommendation. This approach explores both the online and offline social interactions and fuses the content of events to model their joint effect on users' decision-making for events. Extensive experiments are conducted to evaluate the performance of the proposed model on Meetup dataset. The experimental results demonstrate that the proposed model outperforms state-of-the-art methods.
2019,30(3):822-844
, DOI: 10.13328/j.cnki.jos.005681
Abstract:
Since the factorization machine (FM) model can effectively solve the sparsity problem of high-dimensional data feature combination with high prediction accuracy and computational efficiency, it has been widely studied and applied in the field of click-through-rate (CTR) prediction and recommender systems. The review of the progress on the subsequent research on FM and its related models will help to promote the further improvement and application of the model. By comparing the relationship between the FM model and the polynomial regression model and the factorization model, the flexibility and generality of the FM model are described. Considering width extension, the strategies, methods, and key technologies are summarized from the dimensions of high-order feature interaction, field-aware feature interaction and hierarchical feature interaction, as well as feature extraction, combining, intelligent selection and promotion based on feature engineering. The integration approaches and benefits of FM model with other models, especially the combination with deep learning models are compared and analyzed, which provides insights into the in-depth expansion of traditional models. The learning and optimization methods of FM models and the implementation based on different parallel and distributed computing frameworks are summarized, compared, and analyzed. Finally, the authors forecast the difficult points, hot spots and development trends in the FM model that need to be further studied.
Since the factorization machine (FM) model can effectively solve the sparsity problem of high-dimensional data feature combination with high prediction accuracy and computational efficiency, it has been widely studied and applied in the field of click-through-rate (CTR) prediction and recommender systems. The review of the progress on the subsequent research on FM and its related models will help to promote the further improvement and application of the model. By comparing the relationship between the FM model and the polynomial regression model and the factorization model, the flexibility and generality of the FM model are described. Considering width extension, the strategies, methods, and key technologies are summarized from the dimensions of high-order feature interaction, field-aware feature interaction and hierarchical feature interaction, as well as feature extraction, combining, intelligent selection and promotion based on feature engineering. The integration approaches and benefits of FM model with other models, especially the combination with deep learning models are compared and analyzed, which provides insights into the in-depth expansion of traditional models. The learning and optimization methods of FM models and the implementation based on different parallel and distributed computing frameworks are summarized, compared, and analyzed. Finally, the authors forecast the difficult points, hot spots and development trends in the FM model that need to be further studied.
2018,29(12):3747-3763
, DOI: 10.13328/j.cnki.jos.005322
Abstract:
The development of Internet has brought convenience to the public, but also troubles users in making choices among enormous data. Thus, recommender systems based on user understanding are urgently in need. Different from the traditional techniques that usually focus on individual users, the social-based recommender systems perform better with integrating social influence modeling to achieve more accurate user profiling. However, current works usually generalize influence in simple mode, while deep discussions on intrinsic mechanism have been largely ignored. To solve this problem, this paper studies the social influence within users who affects both rating and user attributes, and then proposes a novel trust-driven PMF (TPMF) algorithm to merge these two mechanisms. Furthermore, to deal with the task that different user should have personalized parameters, the study clusters users according to rating correlation and then maps them to corresponding weights, thereby achieving the personalized selection of users' model parameters. Comprehensive experiments on open data sets validate that TPMF and its derivation algorithm can effectively predict users' rating compared with several state of the art baselines, which demonstrates the capability of the presented influence mechanism and technical framework.
The development of Internet has brought convenience to the public, but also troubles users in making choices among enormous data. Thus, recommender systems based on user understanding are urgently in need. Different from the traditional techniques that usually focus on individual users, the social-based recommender systems perform better with integrating social influence modeling to achieve more accurate user profiling. However, current works usually generalize influence in simple mode, while deep discussions on intrinsic mechanism have been largely ignored. To solve this problem, this paper studies the social influence within users who affects both rating and user attributes, and then proposes a novel trust-driven PMF (TPMF) algorithm to merge these two mechanisms. Furthermore, to deal with the task that different user should have personalized parameters, the study clusters users according to rating correlation and then maps them to corresponding weights, thereby achieving the personalized selection of users' model parameters. Comprehensive experiments on open data sets validate that TPMF and its derivation algorithm can effectively predict users' rating compared with several state of the art baselines, which demonstrates the capability of the presented influence mechanism and technical framework.
2019,30(11):3397-3412
, DOI: 10.13328/j.cnki.jos.005545
Abstract:
Recommending valuable and interesting contents for microblog users is an important way to improve the user experience. In this study, tags are considered as the users' interests and a microblog recommendation method based on hypergraph random walk tag Extension and tag probability correlation is proposed via the analysis of characteristics and the existing limitations of microblog recommendation algorithm. Firstly, microblogs are considered as hyperedges, while each term is taken as the hypervertex, and the weighting strategies for both hyperedges and hypervertexes are established. A random walk is conducted on the hypergraph to obtain a number of keywords for the expansion of microblog users. And then the weight of the tag for each user is enhanced based on the relevance weighting scheme and the user tag matrix can be constructed. Probability correlation between tags is calculated to construct the tag similarity matrix, which can be used to update the matrix is updated using the label similarity matrix, which contains both the user interest information and the relationship between tags and tags. Experimental results show that the algorithm is effective in microblog recommendation.
Recommending valuable and interesting contents for microblog users is an important way to improve the user experience. In this study, tags are considered as the users' interests and a microblog recommendation method based on hypergraph random walk tag Extension and tag probability correlation is proposed via the analysis of characteristics and the existing limitations of microblog recommendation algorithm. Firstly, microblogs are considered as hyperedges, while each term is taken as the hypervertex, and the weighting strategies for both hyperedges and hypervertexes are established. A random walk is conducted on the hypergraph to obtain a number of keywords for the expansion of microblog users. And then the weight of the tag for each user is enhanced based on the relevance weighting scheme and the user tag matrix can be constructed. Probability correlation between tags is calculated to construct the tag similarity matrix, which can be used to update the matrix is updated using the label similarity matrix, which contains both the user interest information and the relationship between tags and tags. Experimental results show that the algorithm is effective in microblog recommendation.
2020,31(2):421-438
, DOI: 10.13328/j.cnki.jos.005618
Abstract:
The newly emerging event-based social network (EBSN) based on the event as the core combines the online relationship with offline activities to promote the formation of real and effective social relationship among users. However, excessive activity information would make users difficult to distinguish and choose. The context-aware local event recommendation is an effective solution for the information overload problem, but most of existing local event recommendation algorithms only learns users' preference for contextual information indirectly from statistics of historical event participation and ignores latent correlations among them, which impacts on recommendation effectiveness. To take full advantage of latent correlations between users' event preference and contextual information, the proposed collective contextual relation learning (CCRL) algorithm models relations among users' participation records and related contextual information such as event organizer, description text, venue, and starting time. Then multi-relational Bayesian personalized ranking (MRBPR) algorithm is adapted for collective contextual relation learning and local event recommendation. Experiment results on Meetup dataset demonstrate that proposed algorithm outperforms state-of-the-art local event recommendation algorithms in terms of many metrics.
The newly emerging event-based social network (EBSN) based on the event as the core combines the online relationship with offline activities to promote the formation of real and effective social relationship among users. However, excessive activity information would make users difficult to distinguish and choose. The context-aware local event recommendation is an effective solution for the information overload problem, but most of existing local event recommendation algorithms only learns users' preference for contextual information indirectly from statistics of historical event participation and ignores latent correlations among them, which impacts on recommendation effectiveness. To take full advantage of latent correlations between users' event preference and contextual information, the proposed collective contextual relation learning (CCRL) algorithm models relations among users' participation records and related contextual information such as event organizer, description text, venue, and starting time. Then multi-relational Bayesian personalized ranking (MRBPR) algorithm is adapted for collective contextual relation learning and local event recommendation. Experiment results on Meetup dataset demonstrate that proposed algorithm outperforms state-of-the-art local event recommendation algorithms in terms of many metrics.
2015,26(1):145-166
, DOI: 10.13328/j.cnki.jos.004688
Abstract:
The explosive growth of the digital data brings great challenges to the relational database management systems in addressing issues in areas such as scalability and fault tolerance. The cloud computing techniques have been widely used in many applications and become the standard effective approach to manage large scale data because of their high scalability, high availability and fault tolerance. The existing cloud-based data management systems can't efficiently support complex queries such as multi-dimensional queries and join queries because of lacking of index or view techniques, limiting the application of cloud computing in many respects. This paper conducts an in-depth research on the index techniques for cloud data management to highlight their strengths and weaknesses. This paper also introduces its own preliminary work on the index for massive IOT data in cloud environment. Finally, it points out some challenges in the index techniques for big data in cloud environment.
The explosive growth of the digital data brings great challenges to the relational database management systems in addressing issues in areas such as scalability and fault tolerance. The cloud computing techniques have been widely used in many applications and become the standard effective approach to manage large scale data because of their high scalability, high availability and fault tolerance. The existing cloud-based data management systems can't efficiently support complex queries such as multi-dimensional queries and join queries because of lacking of index or view techniques, limiting the application of cloud computing in many respects. This paper conducts an in-depth research on the index techniques for cloud data management to highlight their strengths and weaknesses. This paper also introduces its own preliminary work on the index for massive IOT data in cloud environment. Finally, it points out some challenges in the index techniques for big data in cloud environment.
2005,16(1):1-7
Abstract:
The paper gives some thinking according to the following four aspects: 1) from the law of things development, revealing the development history of software engineering technology; 2) from the point of software natural characteristic, analyzing the construction of every abstraction layer of virtual machine; 3) from the point of software development, proposing the research content of software engineering discipline, and research the pattern of industrialized software production; 4) based on the appearance of Internet technology, exploring the development trend of software technology.
The paper gives some thinking according to the following four aspects: 1) from the law of things development, revealing the development history of software engineering technology; 2) from the point of software natural characteristic, analyzing the construction of every abstraction layer of virtual machine; 3) from the point of software development, proposing the research content of software engineering discipline, and research the pattern of industrialized software production; 4) based on the appearance of Internet technology, exploring the development trend of software technology.
2012,23(1):1-20
, DOI: 10.3724/SP.J.1001.2012.04100
Abstract:
Context-Aware recommender systems, aiming to further improve performance accuracy and user satisfaction by fully utilizing contextual information, have recently become one of the hottest topics in the domain of recommender systems. This paper presents an overview of the field of context-aware recommender systems from a process-oriented perspective, including system frameworks, key techniques, main models, evaluation, and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
Context-Aware recommender systems, aiming to further improve performance accuracy and user satisfaction by fully utilizing contextual information, have recently become one of the hottest topics in the domain of recommender systems. This paper presents an overview of the field of context-aware recommender systems from a process-oriented perspective, including system frameworks, key techniques, main models, evaluation, and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
2010,21(8):1834-1848
Abstract:
This paper surveys the state of the art of sentiment analysis. First, three important tasks of sentiment analysis are summarized and analyzed in detail, including sentiment extraction, sentiment classification, sentiment retrieval and summarization. Then, the evaluation and corpus for sentiment analysis are introduced. Finally, the applications of sentiment analysis are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, making detailed comparison and analysis.
This paper surveys the state of the art of sentiment analysis. First, three important tasks of sentiment analysis are summarized and analyzed in detail, including sentiment extraction, sentiment classification, sentiment retrieval and summarization. Then, the evaluation and corpus for sentiment analysis are introduced. Finally, the applications of sentiment analysis are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, making detailed comparison and analysis.
2004,15(3):428-442
Abstract:
With the rapid development of e-business, web applications based on the Web are developed from localization to globalization, from B2C(business-to-customer) to B2B(business-to-business), from centralized fashion to decentralized fashion. Web service is a new application model for decentralized computing, and it is also an effective mechanism for the data and service integration on the web. Thus, web service has become a solution to e-business. It is important and necessary to carry out the research on the new architecture of web services, on the combinations with other good techniques, and on the integration of services. In this paper, a survey presents on various aspects of the research of web services from the basic concepts to the principal research problems and the underlying techniques, including data integration in web services, web service composition, semantic web service, web service discovery, web service security, the solution to web services in the P2P (Peer-to-Peer) computing environment, and the grid service, etc. This paper also presents a summary of the current art of the state of these techniques, a discussion on the future research topics, and the challenges of the web services.
With the rapid development of e-business, web applications based on the Web are developed from localization to globalization, from B2C(business-to-customer) to B2B(business-to-business), from centralized fashion to decentralized fashion. Web service is a new application model for decentralized computing, and it is also an effective mechanism for the data and service integration on the web. Thus, web service has become a solution to e-business. It is important and necessary to carry out the research on the new architecture of web services, on the combinations with other good techniques, and on the integration of services. In this paper, a survey presents on various aspects of the research of web services from the basic concepts to the principal research problems and the underlying techniques, including data integration in web services, web service composition, semantic web service, web service discovery, web service security, the solution to web services in the P2P (Peer-to-Peer) computing environment, and the grid service, etc. This paper also presents a summary of the current art of the state of these techniques, a discussion on the future research topics, and the challenges of the web services.
2009,20(1):54-66
Abstract:
Network community structure is one of the most fundamental and important topological properties of complex networks, within which the links between nodes are very dense, but between which they are quite sparse. Network clustering algorithms which aim to discover all natural network communities from given complex networks are fundamentally important for both theoretical researches and practical applications, and can be used to analyze the topological structures, understand the functions, recognize the hidden patterns, and predict the behaviors of complex networks including social networks, biological networks, World Wide Webs and so on. This paper reviews the background, the motivation, the state of arts as well as the main issues of existing works related to discovering network communities, and tries to draw a comprehensive and clear outline for this new and active research area. This work is hopefully beneficial to the researchers from the communities of complex network analysis, data mining, intelligent Web and bioinformatics.
Network community structure is one of the most fundamental and important topological properties of complex networks, within which the links between nodes are very dense, but between which they are quite sparse. Network clustering algorithms which aim to discover all natural network communities from given complex networks are fundamentally important for both theoretical researches and practical applications, and can be used to analyze the topological structures, understand the functions, recognize the hidden patterns, and predict the behaviors of complex networks including social networks, biological networks, World Wide Webs and so on. This paper reviews the background, the motivation, the state of arts as well as the main issues of existing works related to discovering network communities, and tries to draw a comprehensive and clear outline for this new and active research area. This work is hopefully beneficial to the researchers from the communities of complex network analysis, data mining, intelligent Web and bioinformatics.
2017,28(4):860-882
, DOI: 10.13328/j.cnki.jos.005190
Abstract:
Information flow analysis is a promising approach for protecting the confidentiality and integrity of information manipulated by computing systems. Taint analysis, as in practice, is widely used in the area of software security assurance. This survey summarizes the latest advances on taint analysis, especially the solutions applied in different platform applications. Firstly, the basic principle of taint analysis is introduced along with the general technology of taint propagation implemented by dynamic and static analyses. Then, the proposals applied in different platform frameworks, including techniques for protecting privacy leakage on Android and finding security vulnerabilities on Web, are analyzed. Lastly, further research directions and future work are discussed.
Information flow analysis is a promising approach for protecting the confidentiality and integrity of information manipulated by computing systems. Taint analysis, as in practice, is widely used in the area of software security assurance. This survey summarizes the latest advances on taint analysis, especially the solutions applied in different platform applications. Firstly, the basic principle of taint analysis is introduced along with the general technology of taint propagation implemented by dynamic and static analyses. Then, the proposals applied in different platform frameworks, including techniques for protecting privacy leakage on Android and finding security vulnerabilities on Web, are analyzed. Lastly, further research directions and future work are discussed.
2005,16(5):857-868
Abstract:
Wireless Sensor Networks, a novel technology about acquiring and processing information, have been proposed for a multitude of diverse applications. The problem of self-localization, that is, determining where a given node is physically or relatively located in the networks, is a challenging one, and yet extremely crucial for many applications. In this paper, the evaluation criterion of the performance and the taxonomy for wireless sensor networks self-localization systems and algorithms are described, the principles and characteristics of recent representative localization approaches are discussed and presented, and the directions of research in this area are introduced.
Wireless Sensor Networks, a novel technology about acquiring and processing information, have been proposed for a multitude of diverse applications. The problem of self-localization, that is, determining where a given node is physically or relatively located in the networks, is a challenging one, and yet extremely crucial for many applications. In this paper, the evaluation criterion of the performance and the taxonomy for wireless sensor networks self-localization systems and algorithms are described, the principles and characteristics of recent representative localization approaches are discussed and presented, and the directions of research in this area are introduced.
2012,23(4):962-986
, DOI: 10.3724/SP.J.1001.2012.04175
Abstract:
Considered as the next generation computing model, cloud computing plays an important role in scientific and commercial computing area and draws great attention from both academia and industry fields. Under cloud computing environment, data center consist of a large amount of computers, usually up to millions, and stores petabyte even exabyte of data, which may easily lead to the failure of the computers or data. The large amount of computers composition not only leads to great challenges to the scalability of the data center and its storage system, but also results in high hardware infrastructure cost and power cost. Therefore, fault-tolerance, scalability, and power consumption of the distributed storage for a data center becomes key part in the technology of cloud computing, in order to ensure the data availability and reliability. In this paper, a survey is made on the state of art of the key technologies in cloud computing in the following aspects: Design of data center network, organization and arrangement of data, strategies to improve fault-tolerance, methods to save storage space, and energy. Firstly, many kinds of classical topologies of data center network are introduced and compared. Secondly, kinds of current fault-tolerant storage techniques are discussed, and data replication and erasure code strategies are especially compared. Thirdly, the main current energy saving technology is addressed and analyzed. Finally, challenges in distributed storage are reviewed as well as future research trends are predicted.
Considered as the next generation computing model, cloud computing plays an important role in scientific and commercial computing area and draws great attention from both academia and industry fields. Under cloud computing environment, data center consist of a large amount of computers, usually up to millions, and stores petabyte even exabyte of data, which may easily lead to the failure of the computers or data. The large amount of computers composition not only leads to great challenges to the scalability of the data center and its storage system, but also results in high hardware infrastructure cost and power cost. Therefore, fault-tolerance, scalability, and power consumption of the distributed storage for a data center becomes key part in the technology of cloud computing, in order to ensure the data availability and reliability. In this paper, a survey is made on the state of art of the key technologies in cloud computing in the following aspects: Design of data center network, organization and arrangement of data, strategies to improve fault-tolerance, methods to save storage space, and energy. Firstly, many kinds of classical topologies of data center network are introduced and compared. Secondly, kinds of current fault-tolerant storage techniques are discussed, and data replication and erasure code strategies are especially compared. Thirdly, the main current energy saving technology is addressed and analyzed. Finally, challenges in distributed storage are reviewed as well as future research trends are predicted.
2014,25(2):400-418
, DOI: 10.13328/j.cnki.jos.004540
Abstract:
Cyber-Physical Systems (CPSs) have great potentials in several application domains. Time plays an important role in CPS and should be specified in the very early phase of requirements engineering. This paper proposes a framework to model and verify timing requirements for the CPS. To begin with, a conceptual model is presented for providing basic concepts of timing and functional requirements. Guided by this model, the CPS software timing requirement specification can be obtained from CPS environment properties and constraints. To support formal verification, formal semantics for the conceptual model is provided. Based on the semantics, the consistency properties of the timing requirements specification are defined and expressed as CTL formulas. The timing requirements specification is transformed into a NuSMV model and checked by this well-known model checker.
Cyber-Physical Systems (CPSs) have great potentials in several application domains. Time plays an important role in CPS and should be specified in the very early phase of requirements engineering. This paper proposes a framework to model and verify timing requirements for the CPS. To begin with, a conceptual model is presented for providing basic concepts of timing and functional requirements. Guided by this model, the CPS software timing requirement specification can be obtained from CPS environment properties and constraints. To support formal verification, formal semantics for the conceptual model is provided. Based on the semantics, the consistency properties of the timing requirements specification are defined and expressed as CTL formulas. The timing requirements specification is transformed into a NuSMV model and checked by this well-known model checker.
2012,23(1):32-45
, DOI: 10.3724/SP.J.1001.2012.04091
Abstract:
In many areas such as science, simulation, Internet, and e-commerce, the volume of data to be analyzed grows rapidly. Parallel techniques which could be expanded cost-effectively should be invented to deal with the big data. Relational data management technique has gone through a history of nearly 40 years. Now it encounters the tough obstacle of scalability, which relational techniques can not handle large data easily. In the mean time, none relational techniques, such as MapReduce as a typical representation, emerge as a new force, and expand their application from Web search to territories that used to be occupied by relational database systems. They confront relational technique with high availability, high scalability and massive parallel processing capability. Relational technique community, after losing the big deal of Web search, begins to learn from MapReduce. MapReduce also borrows valuable ideas from relational technique community to improve performance. Relational technique and MapReduce compete with each other, and learn from each other; new data analysis platform and new data analysis eco-system are emerging. Finally the two camps of techniques will find their right places in the new eco-system of big data analysis.
In many areas such as science, simulation, Internet, and e-commerce, the volume of data to be analyzed grows rapidly. Parallel techniques which could be expanded cost-effectively should be invented to deal with the big data. Relational data management technique has gone through a history of nearly 40 years. Now it encounters the tough obstacle of scalability, which relational techniques can not handle large data easily. In the mean time, none relational techniques, such as MapReduce as a typical representation, emerge as a new force, and expand their application from Web search to territories that used to be occupied by relational database systems. They confront relational technique with high availability, high scalability and massive parallel processing capability. Relational technique community, after losing the big deal of Web search, begins to learn from MapReduce. MapReduce also borrows valuable ideas from relational technique community to improve performance. Relational technique and MapReduce compete with each other, and learn from each other; new data analysis platform and new data analysis eco-system are emerging. Finally the two camps of techniques will find their right places in the new eco-system of big data analysis.
2010,21(5):899-915
Abstract:
This paper firstly presents a summary of AADL (architecture analysis and design language), including its progress over the years and its modeling elements. Then, it surveys the research and practice of AADL from a model-based perspective, such as AADL modeling, AADL formal semantics, model transformation, verification and code generation. Finally, the potential research directions are discussed.
This paper firstly presents a summary of AADL (architecture analysis and design language), including its progress over the years and its modeling elements. Then, it surveys the research and practice of AADL from a model-based perspective, such as AADL modeling, AADL formal semantics, model transformation, verification and code generation. Finally, the potential research directions are discussed.
2009,20(1):124-137
Abstract:
The appearance of plenty of intelligent devices equipped for short-range wireless communications boosts the fast rise of wireless ad hoc networks application. However, in many realistic application environments, nodes form a disconnected network for most of the time due to nodal mobility, low density, lossy link, etc. Conventional communication model of mobile ad hoc network (MANET) requires at least one path existing from source to destination nodes, which results in communication failure in these scenarios. Opportunistic networks utilize the communication opportunities arising from node movement to forward messages in a hop-by-hop way, and implement communications between nodes based on the "store-carry-forward" routing pattern. This networking approach, totally different from the traditional communication model, captures great interests from researchers. This paper first introduces the conceptions and theories of opportunistic networks and some current typical applications. Then it elaborates the popular research problems including opportunistic forwarding mechanism, mobility model and opportunistic data dissemination and retrieval. Some other interesting research points such as communication middleware, cooperation and security problem and new applications are stated briefly. Finally, the paper concludes and looks forward to the possible research focuses for opportunistic networks in the future.
The appearance of plenty of intelligent devices equipped for short-range wireless communications boosts the fast rise of wireless ad hoc networks application. However, in many realistic application environments, nodes form a disconnected network for most of the time due to nodal mobility, low density, lossy link, etc. Conventional communication model of mobile ad hoc network (MANET) requires at least one path existing from source to destination nodes, which results in communication failure in these scenarios. Opportunistic networks utilize the communication opportunities arising from node movement to forward messages in a hop-by-hop way, and implement communications between nodes based on the "store-carry-forward" routing pattern. This networking approach, totally different from the traditional communication model, captures great interests from researchers. This paper first introduces the conceptions and theories of opportunistic networks and some current typical applications. Then it elaborates the popular research problems including opportunistic forwarding mechanism, mobility model and opportunistic data dissemination and retrieval. Some other interesting research points such as communication middleware, cooperation and security problem and new applications are stated briefly. Finally, the paper concludes and looks forward to the possible research focuses for opportunistic networks in the future.
2009,20(3):524-545
Abstract:
Nowadays it has been widely accepted that the quality of software highly depends on the process that iscarried out in an organization. As part of the effort to support software process engineering activities, the researchon software process modeling and analysis is to provide an effective means to represent and analyze a process and,by doing so, to enhance the understanding of the modeled process. In addition, an enactable process model canprovide a direct guidance for the actual development process. Thus, the enforcement of the process model candirectly contribute to the improvement of the software quality. In this paper, a systematic review is carried out tosurvey the recent development in software process modeling. 72 papers from 20 conference proceedings and 7journals are identified as the evidence. The review aims to promote a better understanding of the literature byanswering the following three questions: 1) What kinds of paradigms are existing methods based on? 2) What kinds of purposes does the existing research have? 3) What kinds of new trends are reflected in the current research? Afterproviding the systematic review, we present our software process modeling method based on a multi-dimensionaland integration methodology that is intended to address several core issues facing the community.
Nowadays it has been widely accepted that the quality of software highly depends on the process that iscarried out in an organization. As part of the effort to support software process engineering activities, the researchon software process modeling and analysis is to provide an effective means to represent and analyze a process and,by doing so, to enhance the understanding of the modeled process. In addition, an enactable process model canprovide a direct guidance for the actual development process. Thus, the enforcement of the process model candirectly contribute to the improvement of the software quality. In this paper, a systematic review is carried out tosurvey the recent development in software process modeling. 72 papers from 20 conference proceedings and 7journals are identified as the evidence. The review aims to promote a better understanding of the literature byanswering the following three questions: 1) What kinds of paradigms are existing methods based on? 2) What kinds of purposes does the existing research have? 3) What kinds of new trends are reflected in the current research? Afterproviding the systematic review, we present our software process modeling method based on a multi-dimensionaland integration methodology that is intended to address several core issues facing the community.
2009,20(2):350-362
Abstract:
This paper makes a comprehensive survey of the recommender system research aiming to facilitate readers to understand this field. First the research background is introduced, including commercial application demands, academic institutes, conferences and journals. After formally and informally describing the recommendation problem, a comparison study is conducted based on categorized algorithms. In addition, the commonly adopted benchmarked datasets and evaluation methods are exhibited and most difficulties and future directions are concluded.
This paper makes a comprehensive survey of the recommender system research aiming to facilitate readers to understand this field. First the research background is introduced, including commercial application demands, academic institutes, conferences and journals. After formally and informally describing the recommendation problem, a comparison study is conducted based on categorized algorithms. In addition, the commonly adopted benchmarked datasets and evaluation methods are exhibited and most difficulties and future directions are concluded.
2004,15(8):1208-1219
Abstract:
With the explosive growth of network applications and complexity, the threat of Internet worms against network security becomes increasingly serious. Especially under the environment of Internet, the variety of the propagation ways and the complexity of the application environment result in worm with much higher frequency of outbreak, much deeper latency and more wider coverage, and Internet worms have been a primary issue faced by malicious code researchers. In this paper, the concept and research situation of Internet worms, exploration function component and execution mechanism are first presented, then the scanning strategies and propagation model are discussed, and finally the critical techniques of Internet worm prevention are given. Some major problems and research trends in this area are also addressed.
With the explosive growth of network applications and complexity, the threat of Internet worms against network security becomes increasingly serious. Especially under the environment of Internet, the variety of the propagation ways and the complexity of the application environment result in worm with much higher frequency of outbreak, much deeper latency and more wider coverage, and Internet worms have been a primary issue faced by malicious code researchers. In this paper, the concept and research situation of Internet worms, exploration function component and execution mechanism are first presented, then the scanning strategies and propagation model are discussed, and finally the critical techniques of Internet worm prevention are given. Some major problems and research trends in this area are also addressed.
2009,20(11):2965-2976
Abstract:
This paper studies uncertain graph data mining and especially investigates the problem of mining frequent subgraph patterns from uncertain graph data. A data model is introduced for representing uncertainties in graphs, and an expected support is employed to evaluate the significance of subgraph patterns. By using the apriori property of expected support, a depth-first search-based mining algorithm is proposed with an efficient method for computing expected supports and a technique for pruning search space, which reduces the number of subgraph isomorphism testings needed by computing expected support from the exponential scale to the linear scale. Experimental results show that the proposed algorithm is 3 to 5 orders of magnitude faster than a na?ve depth-first search algorithm, and is efficient and scalable.
This paper studies uncertain graph data mining and especially investigates the problem of mining frequent subgraph patterns from uncertain graph data. A data model is introduced for representing uncertainties in graphs, and an expected support is employed to evaluate the significance of subgraph patterns. By using the apriori property of expected support, a depth-first search-based mining algorithm is proposed with an efficient method for computing expected supports and a technique for pruning search space, which reduces the number of subgraph isomorphism testings needed by computing expected support from the exponential scale to the linear scale. Experimental results show that the proposed algorithm is 3 to 5 orders of magnitude faster than a na?ve depth-first search algorithm, and is efficient and scalable.
2009,20(5):1226-1240
Abstract:
This paper introduces the concrete details of combining the automated reasoning techniques with planning methods, which includes planning as satisfiability using propositional logic, Conformant planning using modal logic and disjunctive reasoning, planning as nonmonotonic logic, and Flexible planning as fuzzy description logic. After considering experimental results of International Planning Competition and relevant papers, it concludes that planning methods based on automated reasoning techniques is helpful and can be adopted. It also proposes the challenges and possible hotspots.
This paper introduces the concrete details of combining the automated reasoning techniques with planning methods, which includes planning as satisfiability using propositional logic, Conformant planning using modal logic and disjunctive reasoning, planning as nonmonotonic logic, and Flexible planning as fuzzy description logic. After considering experimental results of International Planning Competition and relevant papers, it concludes that planning methods based on automated reasoning techniques is helpful and can be adopted. It also proposes the challenges and possible hotspots.
2003,14(10):1717-1727
Abstract:
Sensor networks are integration of sensor techniques, nested computation techniques, distributed computation techniques and wireless communication techniques. They can be used for testing, sensing, collecting and processing information of monitored objects and transferring the processed information to users. Sensor network is a new research area of computer science and technology and has a wide application future. Both academia and industries are very interested in it. The concepts and characteristics of the sensor networks and the data in the networks are introduced, and the issues of the sensor networks and the data management of sensor networks are discussed. The advance of the research on sensor networks and the data management of sensor networks are also presented.
Sensor networks are integration of sensor techniques, nested computation techniques, distributed computation techniques and wireless communication techniques. They can be used for testing, sensing, collecting and processing information of monitored objects and transferring the processed information to users. Sensor network is a new research area of computer science and technology and has a wide application future. Both academia and industries are very interested in it. The concepts and characteristics of the sensor networks and the data in the networks are introduced, and the issues of the sensor networks and the data management of sensor networks are discussed. The advance of the research on sensor networks and the data management of sensor networks are also presented.
2014,25(4):839-862
, DOI: 10.13328/j.cnki.jos.004558
Abstract:
Batch computing and stream computing are two important forms of big data computing. The research and discussions on batch computing in big data environment are comparatively sufficient. But how to efficiently deal with stream computing to meet many requirements, such as low latency, high throughput and continuously reliable running, and how to build efficient stream big data computing systems, are great challenges in the big data computing research. This paper provides a research of the data computing architecture and the key issues in stream computing in big data environments. Firstly, the research gives a brief summary of three application scenarios of stream computing in business intelligence, marketing and public service. It also shows distinctive features of the stream computing in big data environment, such as real time, volatility, burstiness, irregularity and infinity. A well-designed stream computing system always optimizes in system structure, data transmission, application interfaces, high-availability, and so on. Subsequently, the research offers detailed analyses and comparisons of five typical and open-source stream computing systems in big data environment. Finally, the research specifically addresses some new challenges of the stream big data systems, such as scalability, fault tolerance, consistency, load balancing and throughput.
Batch computing and stream computing are two important forms of big data computing. The research and discussions on batch computing in big data environment are comparatively sufficient. But how to efficiently deal with stream computing to meet many requirements, such as low latency, high throughput and continuously reliable running, and how to build efficient stream big data computing systems, are great challenges in the big data computing research. This paper provides a research of the data computing architecture and the key issues in stream computing in big data environments. Firstly, the research gives a brief summary of three application scenarios of stream computing in business intelligence, marketing and public service. It also shows distinctive features of the stream computing in big data environment, such as real time, volatility, burstiness, irregularity and infinity. A well-designed stream computing system always optimizes in system structure, data transmission, application interfaces, high-availability, and so on. Subsequently, the research offers detailed analyses and comparisons of five typical and open-source stream computing systems in big data environment. Finally, the research specifically addresses some new challenges of the stream big data systems, such as scalability, fault tolerance, consistency, load balancing and throughput.
2000,11(11):1460-1466
Abstract:
Intrusion detection is a highlighted topic of network security research in recent years. In this paper, first the necessity o f intrusion detection is presented, and its concepts and models are described. T hen, many intrusion detection techniques and architectures are summarized. Final ly, the existing problems and the future direction in this field are discussed.
Intrusion detection is a highlighted topic of network security research in recent years. In this paper, first the necessity o f intrusion detection is presented, and its concepts and models are described. T hen, many intrusion detection techniques and architectures are summarized. Final ly, the existing problems and the future direction in this field are discussed.
2012,23(5):1148-1166
, DOI: 10.3724/SP.J.1001.2012.04195
Abstract:
With the recent development of cloud computing, the importance of cloud databases has been widely acknowledged. Here, the features, influence and related products of cloud databases are first discussed. Then, research issues of cloud databases are presented in detail, which include data model, architecture, consistency, programming model, data security, performance optimization, benchmark, and so on. Finally, some future trends in this area are discussed.
With the recent development of cloud computing, the importance of cloud databases has been widely acknowledged. Here, the features, influence and related products of cloud databases are first discussed. Then, research issues of cloud databases are presented in detail, which include data model, architecture, consistency, programming model, data security, performance optimization, benchmark, and so on. Finally, some future trends in this area are discussed.
2013,24(8):1786-1803
, DOI: 10.3724/SP.J.1001.2013.04416
Abstract:
Many specific application oriented NoSQL database systems are developed for satisfying the new requirement of big data management. This paper surveys researches on typical NoSQL database based on key-value data model. First, the characteristics of big data, and the key technique issues supporting big data management are introduced. Then frontier efforts and research challenges are given, including system architecture, data model, access mode, index, transaction, system elasticity, load balance, replica strategy, data consistency, flash cache, MapReduce based data process and new generation data management system etc. Finally, research prospects are given.
Many specific application oriented NoSQL database systems are developed for satisfying the new requirement of big data management. This paper surveys researches on typical NoSQL database based on key-value data model. First, the characteristics of big data, and the key technique issues supporting big data management are introduced. Then frontier efforts and research challenges are given, including system architecture, data model, access mode, index, transaction, system elasticity, load balance, replica strategy, data consistency, flash cache, MapReduce based data process and new generation data management system etc. Finally, research prospects are given.
2009,20(10):2729-2743
Abstract:
In a multi-hop wireless sensor network (WSN), the sensors closest to the sink tend to deplete their energy faster than other sensors, which is known as an energy hole around the sink. No more data can be delivered to the sink after an energy hole appears, while a considerable amount of energy is wasted and the network lifetime ends prematurely. This paper investigates the energy hole problem, and based on the improved corona model with levels, it concludes that the assignment of transmission ranges of nodes in different coronas is an effective approach for achieving energy-efficient network. It proves that the optimal transmission ranges for all areas is a multi-objective optimization problem (MOP), which is NP hard. The paper proposes an ACO (ant colony optimization)-based distributed algorithm to prolong the network lifetime, which can help nodes in different areas to adaptively find approximate optimal transmission range based on the node distribution. Furthermore, the simulation results indicate that the network lifetime under this solution approximates to that using the optimal list. Compared with existing algorithms, this ACO-based algorithm can not only make the network lifetime be extended more than two times longer, but also have good performance in the non-uniform node distribution.
In a multi-hop wireless sensor network (WSN), the sensors closest to the sink tend to deplete their energy faster than other sensors, which is known as an energy hole around the sink. No more data can be delivered to the sink after an energy hole appears, while a considerable amount of energy is wasted and the network lifetime ends prematurely. This paper investigates the energy hole problem, and based on the improved corona model with levels, it concludes that the assignment of transmission ranges of nodes in different coronas is an effective approach for achieving energy-efficient network. It proves that the optimal transmission ranges for all areas is a multi-objective optimization problem (MOP), which is NP hard. The paper proposes an ACO (ant colony optimization)-based distributed algorithm to prolong the network lifetime, which can help nodes in different areas to adaptively find approximate optimal transmission range based on the node distribution. Furthermore, the simulation results indicate that the network lifetime under this solution approximates to that using the optimal list. Compared with existing algorithms, this ACO-based algorithm can not only make the network lifetime be extended more than two times longer, but also have good performance in the non-uniform node distribution.
2006,17(7):1588-1600
Abstract:
Routing technology at the network layer is pivotal in the architecture of wireless sensor networks. As an active branch of routing technology, cluster-based routing protocols excel in network topology management, energy minimization, data aggregation and so on. In this paper, cluster-based routing mechanisms for wireless sensor networks are analyzed. Cluster head selection, cluster formation and data transmission are three key techniques in cluster-based routing protocols. As viewed from the three techniques, recent representative cluster-based routing protocols are presented, and their characteristics and application areas are compared. Finally, the future research issues in this area are pointed out.
Routing technology at the network layer is pivotal in the architecture of wireless sensor networks. As an active branch of routing technology, cluster-based routing protocols excel in network topology management, energy minimization, data aggregation and so on. In this paper, cluster-based routing mechanisms for wireless sensor networks are analyzed. Cluster head selection, cluster formation and data transmission are three key techniques in cluster-based routing protocols. As viewed from the three techniques, recent representative cluster-based routing protocols are presented, and their characteristics and application areas are compared. Finally, the future research issues in this area are pointed out.
2002,13(7):1228-1237
Abstract:
Software architecture (SA) is emerging as one of the primary research areas in software engineering recently and one of the key technologies to the development of large-scale software-intensive system and software product line system. The history and the major direction of SA are summarized, and the concept of SA is brought up based on analyzing and comparing the several classical definitions about SA. Based on summing up the activities about SA, two categories of study about SA are extracted out, and the advancements of researches on SA are subsequently introduced from seven aspects.Additionally,some disadvantages of study on SA are discussed,and the causes are explained at the same.Finally,it is concluded with some singificantly promising tendency about research on SA.
Software architecture (SA) is emerging as one of the primary research areas in software engineering recently and one of the key technologies to the development of large-scale software-intensive system and software product line system. The history and the major direction of SA are summarized, and the concept of SA is brought up based on analyzing and comparing the several classical definitions about SA. Based on summing up the activities about SA, two categories of study about SA are extracted out, and the advancements of researches on SA are subsequently introduced from seven aspects.Additionally,some disadvantages of study on SA are discussed,and the causes are explained at the same.Finally,it is concluded with some singificantly promising tendency about research on SA.
-
全文下载排行(总排行，年度排行，各期排行)
摘要点击排行(总排行，年度排行，各期排行)
Article Search
Search by issue
Select AllDeselectExport
Display Method:
2003,14(7):1282-1291
Abstract:
Sensor network, which is made by the convergence of sensor, micro-electro-mechanism system and networks technologies, is a novel technology about acquiring and processing information. In this paper, the architecture of wireless sensor network is briefly introduced. Next, some valuable applications are explained and forecasted. Combining with the existing work, the hot spots including power-aware routing and media access control schemes are discussed and presented in detail. Finally, taking account of application requirements, several future research directions are put forward.
Sensor network, which is made by the convergence of sensor, micro-electro-mechanism system and networks technologies, is a novel technology about acquiring and processing information. In this paper, the architecture of wireless sensor network is briefly introduced. Next, some valuable applications are explained and forecasted. Combining with the existing work, the hot spots including power-aware routing and media access control schemes are discussed and presented in detail. Finally, taking account of application requirements, several future research directions are put forward.
2008,19(1):48-61
Abstract:
The research actuality and new progress in clustering algorithm in recent years are summarized in this paper. First, the analysis and induction of some representative clustering algorithms have been made from several aspects, such as the ideas of algorithm, key technology, advantage and disadvantage. On the other hand, several typical clustering algorithms and known data sets are selected, simulation experiments are implemented from both sides of accuracy and running efficiency, and clustering condition of one algorithm with different data sets is analyzed by comparing with the same clustering of the data set under different algorithms. Finally, the research hotspot, difficulty, shortage of the data clustering and some pending problems are addressed by the integration of the aforementioned two aspects information. The above work can give a valuable reference for data clustering and data mining.
The research actuality and new progress in clustering algorithm in recent years are summarized in this paper. First, the analysis and induction of some representative clustering algorithms have been made from several aspects, such as the ideas of algorithm, key technology, advantage and disadvantage. On the other hand, several typical clustering algorithms and known data sets are selected, simulation experiments are implemented from both sides of accuracy and running efficiency, and clustering condition of one algorithm with different data sets is analyzed by comparing with the same clustering of the data set under different algorithms. Finally, the research hotspot, difficulty, shortage of the data clustering and some pending problems are addressed by the integration of the aforementioned two aspects information. The above work can give a valuable reference for data clustering and data mining.
2011,22(1):71-83
, DOI: 10.3724/SP.J.1001.2011.03958
Abstract:
Cloud Computing is the fundamental change happening in the field of Information Technology. It is a representation of a movement towards the intensive, large scale specialization. On the other hand, it brings about not only convenience and efficiency problems, but also great challenges in the field of data security and privacy protection. Currently, security has been regarded as one of the greatest problems in the development of Cloud Computing. This paper describes the great requirements in Cloud Computing, security key technology, standard and regulation etc., and provides a Cloud Computing security framework. This paper argues that the changes in the above aspects will result in a technical revolution in the field of information security.
Cloud Computing is the fundamental change happening in the field of Information Technology. It is a representation of a movement towards the intensive, large scale specialization. On the other hand, it brings about not only convenience and efficiency problems, but also great challenges in the field of data security and privacy protection. Currently, security has been regarded as one of the greatest problems in the development of Cloud Computing. This paper describes the great requirements in Cloud Computing, security key technology, standard and regulation etc., and provides a Cloud Computing security framework. This paper argues that the changes in the above aspects will result in a technical revolution in the field of information security.
2010,21(8):1834-1848
Abstract:
This paper surveys the state of the art of sentiment analysis. First, three important tasks of sentiment analysis are summarized and analyzed in detail, including sentiment extraction, sentiment classification, sentiment retrieval and summarization. Then, the evaluation and corpus for sentiment analysis are introduced. Finally, the applications of sentiment analysis are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, making detailed comparison and analysis.
This paper surveys the state of the art of sentiment analysis. First, three important tasks of sentiment analysis are summarized and analyzed in detail, including sentiment extraction, sentiment classification, sentiment retrieval and summarization. Then, the evaluation and corpus for sentiment analysis are introduced. Finally, the applications of sentiment analysis are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, making detailed comparison and analysis.
2009,20(1):54-66
Abstract:
Network community structure is one of the most fundamental and important topological properties of complex networks, within which the links between nodes are very dense, but between which they are quite sparse. Network clustering algorithms which aim to discover all natural network communities from given complex networks are fundamentally important for both theoretical researches and practical applications, and can be used to analyze the topological structures, understand the functions, recognize the hidden patterns, and predict the behaviors of complex networks including social networks, biological networks, World Wide Webs and so on. This paper reviews the background, the motivation, the state of arts as well as the main issues of existing works related to discovering network communities, and tries to draw a comprehensive and clear outline for this new and active research area. This work is hopefully beneficial to the researchers from the communities of complex network analysis, data mining, intelligent Web and bioinformatics.
Network community structure is one of the most fundamental and important topological properties of complex networks, within which the links between nodes are very dense, but between which they are quite sparse. Network clustering algorithms which aim to discover all natural network communities from given complex networks are fundamentally important for both theoretical researches and practical applications, and can be used to analyze the topological structures, understand the functions, recognize the hidden patterns, and predict the behaviors of complex networks including social networks, biological networks, World Wide Webs and so on. This paper reviews the background, the motivation, the state of arts as well as the main issues of existing works related to discovering network communities, and tries to draw a comprehensive and clear outline for this new and active research area. This work is hopefully beneficial to the researchers from the communities of complex network analysis, data mining, intelligent Web and bioinformatics.
2009,20(2):271-289
Abstract:
Evolutionary multi-objective optimization (EMO), whose main task is to deal with multi-objective optimization problems by evolutionary computation, has become a hot topic in evolutionary computation community. After summarizing the EMO algorithms before 2003 briefly, the recent advances in EMO are discussed in details. The current research directions are concluded. On the one hand, more new evolutionary paradigms have been introduced into EMO community, such as particle swarm optimization, artificial immune systems, and estimation distribution algorithms. On the other hand, in order to deal with many-objective optimization problems, many new dominance schemes different from traditional Pareto-dominance come forth. Furthermore, the essential characteristics of multi-objective optimization problems are deeply investigated. This paper also gives experimental comparison of several representative algorithms. Finally, several viewpoints for the future research of EMO are proposed.
Evolutionary multi-objective optimization (EMO), whose main task is to deal with multi-objective optimization problems by evolutionary computation, has become a hot topic in evolutionary computation community. After summarizing the EMO algorithms before 2003 briefly, the recent advances in EMO are discussed in details. The current research directions are concluded. On the one hand, more new evolutionary paradigms have been introduced into EMO community, such as particle swarm optimization, artificial immune systems, and estimation distribution algorithms. On the other hand, in order to deal with many-objective optimization problems, many new dominance schemes different from traditional Pareto-dominance come forth. Furthermore, the essential characteristics of multi-objective optimization problems are deeply investigated. This paper also gives experimental comparison of several representative algorithms. Finally, several viewpoints for the future research of EMO are proposed.
2009,20(5):1337-1348
Abstract:
This paper surveys the current technologies adopted in cloud computing as well as the systems in enterprises. Cloud computing can be viewed from two different aspects. One is about the cloud infrastructure which is the building block for the up layer cloud application. The other is of course the cloud application. This paper focuses on the cloud infrastructure including the systems and current research. Some attractive cloud applications are also discussed. Cloud computing infrastructure has three distinct characteristics. First, the infrastructure is built on top of large scale clusters which contain a large number of cheap PC servers. Second, the applications are co-designed with the fundamental infrastructure that the computing resources can be maximally utilized. Third, the reliability of the whole system is achieved by software building on top of redundant hardware instead of mere hardware. All these technologies are for the two important goals for distributed system: high scalability and high availability. Scalability means that the cloud infrastructure can be expanded to very large scale even to thousands of nodes. Availability means that the services are available even when quite a number of nodes fail. From this paper, readers will capture the current status of cloud computing as well as its future trends.
This paper surveys the current technologies adopted in cloud computing as well as the systems in enterprises. Cloud computing can be viewed from two different aspects. One is about the cloud infrastructure which is the building block for the up layer cloud application. The other is of course the cloud application. This paper focuses on the cloud infrastructure including the systems and current research. Some attractive cloud applications are also discussed. Cloud computing infrastructure has three distinct characteristics. First, the infrastructure is built on top of large scale clusters which contain a large number of cheap PC servers. Second, the applications are co-designed with the fundamental infrastructure that the computing resources can be maximally utilized. Third, the reliability of the whole system is achieved by software building on top of redundant hardware instead of mere hardware. All these technologies are for the two important goals for distributed system: high scalability and high availability. Scalability means that the cloud infrastructure can be expanded to very large scale even to thousands of nodes. Availability means that the services are available even when quite a number of nodes fail. From this paper, readers will capture the current status of cloud computing as well as its future trends.
2014,25(9):1889-1908
, DOI: 10.13328/j.cnki.jos.004674
Abstract:
This paper first introduces the key features of big data in different processing modes and their typical application scenarios, as well as corresponding representative processing systems. It then summarizes three development trends of big data processing systems. Next, the paper gives a brief survey on system supported analytic technologies and applications (including deep learning, knowledge computing, social computing, and visualization), and summarizes the key roles of individual technologies in big data analysis and understanding. Finally, the paper lays out three grand challenges of big data processing and analysis, i.e., data complexity, computation complexity, and system complexity. Potential ways for dealing with each complexity are also discussed.
This paper first introduces the key features of big data in different processing modes and their typical application scenarios, as well as corresponding representative processing systems. It then summarizes three development trends of big data processing systems. Next, the paper gives a brief survey on system supported analytic technologies and applications (including deep learning, knowledge computing, social computing, and visualization), and summarizes the key roles of individual technologies in big data analysis and understanding. Finally, the paper lays out three grand challenges of big data processing and analysis, i.e., data complexity, computation complexity, and system complexity. Potential ways for dealing with each complexity are also discussed.
2010,21(3):427-437
Abstract:
Automatic generation of poetry has always been considered a hard nut in natural language generation.This paper reports some pioneering research on a possible generic algorithm and its automatic generation of SONGCI. In light of the characteristics of Chinese ancient poetry, this paper designed the level and oblique tones-based coding method, the syntactic and semantic weighted function of fitness, the elitism and roulette-combined selection operator, and the partially mapped crossover operator and the heuristic mutation operator. As shown by tests, the system constructed on the basis of the computing model designed in this paper is basically capable of generating Chinese SONGCI with some aesthetic merit. This work represents progress in the field of Chinese poetry automatic generation.
Automatic generation of poetry has always been considered a hard nut in natural language generation.This paper reports some pioneering research on a possible generic algorithm and its automatic generation of SONGCI. In light of the characteristics of Chinese ancient poetry, this paper designed the level and oblique tones-based coding method, the syntactic and semantic weighted function of fitness, the elitism and roulette-combined selection operator, and the partially mapped crossover operator and the heuristic mutation operator. As shown by tests, the system constructed on the basis of the computing model designed in this paper is basically capable of generating Chinese SONGCI with some aesthetic merit. This work represents progress in the field of Chinese poetry automatic generation.
2009,20(2):350-362
Abstract:
This paper makes a comprehensive survey of the recommender system research aiming to facilitate readers to understand this field. First the research background is introduced, including commercial application demands, academic institutes, conferences and journals. After formally and informally describing the recommendation problem, a comparison study is conducted based on categorized algorithms. In addition, the commonly adopted benchmarked datasets and evaluation methods are exhibited and most difficulties and future directions are concluded.
This paper makes a comprehensive survey of the recommender system research aiming to facilitate readers to understand this field. First the research background is introduced, including commercial application demands, academic institutes, conferences and journals. After formally and informally describing the recommendation problem, a comparison study is conducted based on categorized algorithms. In addition, the commonly adopted benchmarked datasets and evaluation methods are exhibited and most difficulties and future directions are concluded.
2021,32(2):349-369
, DOI: 10.13328/j.cnki.jos.006138
Abstract:
Few-shot learning is defined as learning models to solve problems from small samples. In recent years, under the trend of training model with big data, machine learning and deep learning have achieved success in many fields. However, in many application scenarios in the real world, there is not a large amount of data or labeled data for model training, and labeling a large number of unlabeled samples will cost a lot of manpower. Therefore, how to use a small number of samples for learning has become a problem that needs to be paid attention to at present. This paper systematically combs the current approaches of few-shot learning. It introduces each kind of corresponding model from the three categories: fine-tune based, data augmentation based, and transfer learning based. Then, the data augmentation based approaches are subdivided into unlabeled data based, data generation based, and feature augmentation based approaches. The transfer learning based approaches are subdivided into metric learning based, meta-learning based, and graph neural network based methods. In the following, the paper summarizes the few-shot datasets and the results in the experiments of the aforementioned models. Next, the paper summarizes the current situation and challenges in few-shot learning. Finally, the future technological development of few-shot learning is prospected.
Few-shot learning is defined as learning models to solve problems from small samples. In recent years, under the trend of training model with big data, machine learning and deep learning have achieved success in many fields. However, in many application scenarios in the real world, there is not a large amount of data or labeled data for model training, and labeling a large number of unlabeled samples will cost a lot of manpower. Therefore, how to use a small number of samples for learning has become a problem that needs to be paid attention to at present. This paper systematically combs the current approaches of few-shot learning. It introduces each kind of corresponding model from the three categories: fine-tune based, data augmentation based, and transfer learning based. Then, the data augmentation based approaches are subdivided into unlabeled data based, data generation based, and feature augmentation based approaches. The transfer learning based approaches are subdivided into metric learning based, meta-learning based, and graph neural network based methods. In the following, the paper summarizes the few-shot datasets and the results in the experiments of the aforementioned models. Next, the paper summarizes the current situation and challenges in few-shot learning. Finally, the future technological development of few-shot learning is prospected.
2016,27(1):45-71
, DOI: 10.13328/j.cnki.jos.004914
Abstract:
Android is a modern and most popular software platform for smartphones. According to report, Android accounted for a huge 81% of all smartphones in 2014 and shipped over 1 billion units worldwide for the first time ever. Apple, Microsoft, Blackberry and Firefox trailed a long way behind. At the same time, increased popularity of the Android smartphones has attracted hackers, leading to massive increase of Android malware applications. This paper summarizes and analyzes the latest advances in Android security from multidimensional perspectives, covering Android architecture, design principles, security mechanisms, major security threats, classification and detection of malware, static and dynamic analyses, machine learning approaches, and security extension proposals.
Android is a modern and most popular software platform for smartphones. According to report, Android accounted for a huge 81% of all smartphones in 2014 and shipped over 1 billion units worldwide for the first time ever. Apple, Microsoft, Blackberry and Firefox trailed a long way behind. At the same time, increased popularity of the Android smartphones has attracted hackers, leading to massive increase of Android malware applications. This paper summarizes and analyzes the latest advances in Android security from multidimensional perspectives, covering Android architecture, design principles, security mechanisms, major security threats, classification and detection of malware, static and dynamic analyses, machine learning approaches, and security extension proposals.
2004,15(10):1493-1504
Abstract:
Graphics processing unit (GPU) has been developing rapidly in recent years at a speed over Moor抯 law, and as a result, various applications associated with computer graphics advance greatly. At the same time, the highly processing power, parallelism and programmability available nowadays on the contemporary GPU provide an ideal platform on which the general-purpose computation could be made. Starting from an introduction to the development history and the architecture of GPU, the technical fundamentals of GPU are described in the paper. Then in the main part of the paper, the development of various applications on general purpose computation on GPU is introduced, and among those applications, fluid dynamics, algebraic computation, database operations, and spectrum analysis are introduced in detail. The experience of our work on fluid dynamics has been also given, and the development of software tools in this area is introduced. Finally, a conclusion is made, and the future development and the new challenge on both hardware and software in this subject are discussed.
Graphics processing unit (GPU) has been developing rapidly in recent years at a speed over Moor抯 law, and as a result, various applications associated with computer graphics advance greatly. At the same time, the highly processing power, parallelism and programmability available nowadays on the contemporary GPU provide an ideal platform on which the general-purpose computation could be made. Starting from an introduction to the development history and the architecture of GPU, the technical fundamentals of GPU are described in the paper. Then in the main part of the paper, the development of various applications on general purpose computation on GPU is introduced, and among those applications, fluid dynamics, algebraic computation, database operations, and spectrum analysis are introduced in detail. The experience of our work on fluid dynamics has been also given, and the development of software tools in this area is introduced. Finally, a conclusion is made, and the future development and the new challenge on both hardware and software in this subject are discussed.
2013,24(11):2476-2497
, DOI: 10.3724/SP.J.1001.2013.04486
Abstract:
Probabilistic graphical models are powerful tools for compactly representing complex probability distributions, efficiently computing (approximate) marginal and conditional distributions, and conveniently learning parameters and hyperparameters in probabilistic models. As a result, they have been widely used in applications that require some sort of automated probabilistic reasoning, such as computer vision and natural language processing, as a formal approach to deal with uncertainty. This paper surveys the basic concepts and key results of representation, inference and learning in probabilistic graphical models, and demonstrates their uses in two important probabilistic models. It also reviews some recent advances in speeding up classic approximate inference algorithms, followed by a discussion of promising research directions.
Probabilistic graphical models are powerful tools for compactly representing complex probability distributions, efficiently computing (approximate) marginal and conditional distributions, and conveniently learning parameters and hyperparameters in probabilistic models. As a result, they have been widely used in applications that require some sort of automated probabilistic reasoning, such as computer vision and natural language processing, as a formal approach to deal with uncertainty. This paper surveys the basic concepts and key results of representation, inference and learning in probabilistic graphical models, and demonstrates their uses in two important probabilistic models. It also reviews some recent advances in speeding up classic approximate inference algorithms, followed by a discussion of promising research directions.
2022,33(7):2464-2481
, DOI: 10.13328/j.cnki.jos.006585
Abstract:
Symbolic propagation methods based on linear abstraction play a significant role in neural network verification. This study proposes the notion of multi-path back-propagation for these methods. Existing methods are viewed as using only a single back-propagation path to calculate the upper and lower bounds of each node in a given neural network, being specific instances of the proposed notion. Leveraging multiple back-propagation paths effectively improves the accuracy of this kind of method. For evaluation, the proposed method is quantitatively compared using multiple back-propagation paths with the state-of-the-art tool DeepPoly on benchmarks ACAS Xu, MNIST, and CIFAR10. The experiment results show that the proposed method achieves significant accuracy improvement while introducing only a low extra time cost. In addition, the multi-path back-propagation method is compared with the Optimized LiRPA based on global optimization, on the dataset MNIST. The results show that the proposed method still has an accuracy advantage.
Symbolic propagation methods based on linear abstraction play a significant role in neural network verification. This study proposes the notion of multi-path back-propagation for these methods. Existing methods are viewed as using only a single back-propagation path to calculate the upper and lower bounds of each node in a given neural network, being specific instances of the proposed notion. Leveraging multiple back-propagation paths effectively improves the accuracy of this kind of method. For evaluation, the proposed method is quantitatively compared using multiple back-propagation paths with the state-of-the-art tool DeepPoly on benchmarks ACAS Xu, MNIST, and CIFAR10. The experiment results show that the proposed method achieves significant accuracy improvement while introducing only a low extra time cost. In addition, the multi-path back-propagation method is compared with the Optimized LiRPA based on global optimization, on the dataset MNIST. The results show that the proposed method still has an accuracy advantage.
2018,29(5):1471-1514
, DOI: 10.13328/j.cnki.jos.005519
Abstract:
Computer aided detection/diagnosis (CAD) can improve the accuracy of diagnosis,reduce false positive,and provide decision supports for doctors.The main purpose of this paper is to analyze the latest development of computer aided diagnosis tools.Focusing on the top four fatal cancer's incidence positions,major recent publications on CAD applications in different medical imaging areas are reviewed in this survey according to different imaging techniques and diseases.Further more,multidimentional analysis is made on the researches from image data sets,algorithms and evaluation methods.Finally,existing problems,research trend and development direction in the field of medical image CAD system are discussed.
Computer aided detection/diagnosis (CAD) can improve the accuracy of diagnosis,reduce false positive,and provide decision supports for doctors.The main purpose of this paper is to analyze the latest development of computer aided diagnosis tools.Focusing on the top four fatal cancer's incidence positions,major recent publications on CAD applications in different medical imaging areas are reviewed in this survey according to different imaging techniques and diseases.Further more,multidimentional analysis is made on the researches from image data sets,algorithms and evaluation methods.Finally,existing problems,research trend and development direction in the field of medical image CAD system are discussed.
2012,23(1):1-20
, DOI: 10.3724/SP.J.1001.2012.04100
Abstract:
Context-Aware recommender systems, aiming to further improve performance accuracy and user satisfaction by fully utilizing contextual information, have recently become one of the hottest topics in the domain of recommender systems. This paper presents an overview of the field of context-aware recommender systems from a process-oriented perspective, including system frameworks, key techniques, main models, evaluation, and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
Context-Aware recommender systems, aiming to further improve performance accuracy and user satisfaction by fully utilizing contextual information, have recently become one of the hottest topics in the domain of recommender systems. This paper presents an overview of the field of context-aware recommender systems from a process-oriented perspective, including system frameworks, key techniques, main models, evaluation, and typical applications. The prospects for future development and suggestions for possible extensions are also discussed.
2020,31(7):2245-2282
, DOI: 10.13328/j.cnki.jos.006037
Abstract:
Ultrasonography is the first choice of imaging examination and preoperative evaluation for thyroid and breast cancer. However, ultrasonic characteristics of benign and malignant nodules are commonly overlapped. The diagnosis heavily relies on operator's experience other than quantitative and stable methods. In recent years, medical imaging analysis based on computer technology has developed rapidly, and a series of landmark breakthroughs have been made, which provides effective decision supports for medical imaging diagnosis. In this work, the research progress of computer vision and image recognition technologies in thyroid and breast ultrasound images is studied. A series of key technologies involved in automatic diagnosis of ultrasound images is the main lines of the work. The major algorithms in recent years are summarized and analyzed, such as ultrasound image preprocessing, lesion localization and segmentation, feature extraction and classification. Moreover, multi-dimensional analysis is made on the algorithms, data sets, and evaluation methods. Finally, existing problems related to automatic analysis of those two kinds of ultrasound imaging are discussed, research trend and development direction in the field of ultrasound images analysis are discussed.
Ultrasonography is the first choice of imaging examination and preoperative evaluation for thyroid and breast cancer. However, ultrasonic characteristics of benign and malignant nodules are commonly overlapped. The diagnosis heavily relies on operator's experience other than quantitative and stable methods. In recent years, medical imaging analysis based on computer technology has developed rapidly, and a series of landmark breakthroughs have been made, which provides effective decision supports for medical imaging diagnosis. In this work, the research progress of computer vision and image recognition technologies in thyroid and breast ultrasound images is studied. A series of key technologies involved in automatic diagnosis of ultrasound images is the main lines of the work. The major algorithms in recent years are summarized and analyzed, such as ultrasound image preprocessing, lesion localization and segmentation, feature extraction and classification. Moreover, multi-dimensional analysis is made on the algorithms, data sets, and evaluation methods. Finally, existing problems related to automatic analysis of those two kinds of ultrasound imaging are discussed, research trend and development direction in the field of ultrasound images analysis are discussed.
2015,26(1):62-81
, DOI: 10.13328/j.cnki.jos.004701
Abstract:
Network abstraction brings about the naissance of software-defined networking. SDN decouples data plane and control plane, and simplifies network management. The paper starts with a discussion on the background in the naissance and developments of SDN, combing its architecture that includes data layer, control layer and application layer. Then their key technologies are elaborated according to the hierarchical architecture of SDN. The characteristics of consistency, availability, and tolerance are especially analyzed. Moreover, latest achievements for profiled scenes are introduced. The future works are summarized in the end.
Network abstraction brings about the naissance of software-defined networking. SDN decouples data plane and control plane, and simplifies network management. The paper starts with a discussion on the background in the naissance and developments of SDN, combing its architecture that includes data layer, control layer and application layer. Then their key technologies are elaborated according to the hierarchical architecture of SDN. The characteristics of consistency, availability, and tolerance are especially analyzed. Moreover, latest achievements for profiled scenes are introduced. The future works are summarized in the end.
2012,23(4):962-986
, DOI: 10.3724/SP.J.1001.2012.04175
Abstract:
Considered as the next generation computing model, cloud computing plays an important role in scientific and commercial computing area and draws great attention from both academia and industry fields. Under cloud computing environment, data center consist of a large amount of computers, usually up to millions, and stores petabyte even exabyte of data, which may easily lead to the failure of the computers or data. The large amount of computers composition not only leads to great challenges to the scalability of the data center and its storage system, but also results in high hardware infrastructure cost and power cost. Therefore, fault-tolerance, scalability, and power consumption of the distributed storage for a data center becomes key part in the technology of cloud computing, in order to ensure the data availability and reliability. In this paper, a survey is made on the state of art of the key technologies in cloud computing in the following aspects: Design of data center network, organization and arrangement of data, strategies to improve fault-tolerance, methods to save storage space, and energy. Firstly, many kinds of classical topologies of data center network are introduced and compared. Secondly, kinds of current fault-tolerant storage techniques are discussed, and data replication and erasure code strategies are especially compared. Thirdly, the main current energy saving technology is addressed and analyzed. Finally, challenges in distributed storage are reviewed as well as future research trends are predicted.
Considered as the next generation computing model, cloud computing plays an important role in scientific and commercial computing area and draws great attention from both academia and industry fields. Under cloud computing environment, data center consist of a large amount of computers, usually up to millions, and stores petabyte even exabyte of data, which may easily lead to the failure of the computers or data. The large amount of computers composition not only leads to great challenges to the scalability of the data center and its storage system, but also results in high hardware infrastructure cost and power cost. Therefore, fault-tolerance, scalability, and power consumption of the distributed storage for a data center becomes key part in the technology of cloud computing, in order to ensure the data availability and reliability. In this paper, a survey is made on the state of art of the key technologies in cloud computing in the following aspects: Design of data center network, organization and arrangement of data, strategies to improve fault-tolerance, methods to save storage space, and energy. Firstly, many kinds of classical topologies of data center network are introduced and compared. Secondly, kinds of current fault-tolerant storage techniques are discussed, and data replication and erasure code strategies are especially compared. Thirdly, the main current energy saving technology is addressed and analyzed. Finally, challenges in distributed storage are reviewed as well as future research trends are predicted.
2012,23(1):32-45
, DOI: 10.3724/SP.J.1001.2012.04091
Abstract:
In many areas such as science, simulation, Internet, and e-commerce, the volume of data to be analyzed grows rapidly. Parallel techniques which could be expanded cost-effectively should be invented to deal with the big data. Relational data management technique has gone through a history of nearly 40 years. Now it encounters the tough obstacle of scalability, which relational techniques can not handle large data easily. In the mean time, none relational techniques, such as MapReduce as a typical representation, emerge as a new force, and expand their application from Web search to territories that used to be occupied by relational database systems. They confront relational technique with high availability, high scalability and massive parallel processing capability. Relational technique community, after losing the big deal of Web search, begins to learn from MapReduce. MapReduce also borrows valuable ideas from relational technique community to improve performance. Relational technique and MapReduce compete with each other, and learn from each other; new data analysis platform and new data analysis eco-system are emerging. Finally the two camps of techniques will find their right places in the new eco-system of big data analysis.
In many areas such as science, simulation, Internet, and e-commerce, the volume of data to be analyzed grows rapidly. Parallel techniques which could be expanded cost-effectively should be invented to deal with the big data. Relational data management technique has gone through a history of nearly 40 years. Now it encounters the tough obstacle of scalability, which relational techniques can not handle large data easily. In the mean time, none relational techniques, such as MapReduce as a typical representation, emerge as a new force, and expand their application from Web search to territories that used to be occupied by relational database systems. They confront relational technique with high availability, high scalability and massive parallel processing capability. Relational technique community, after losing the big deal of Web search, begins to learn from MapReduce. MapReduce also borrows valuable ideas from relational technique community to improve performance. Relational technique and MapReduce compete with each other, and learn from each other; new data analysis platform and new data analysis eco-system are emerging. Finally the two camps of techniques will find their right places in the new eco-system of big data analysis.
2005,16(5):857-868
Abstract:
Wireless Sensor Networks, a novel technology about acquiring and processing information, have been proposed for a multitude of diverse applications. The problem of self-localization, that is, determining where a given node is physically or relatively located in the networks, is a challenging one, and yet extremely crucial for many applications. In this paper, the evaluation criterion of the performance and the taxonomy for wireless sensor networks self-localization systems and algorithms are described, the principles and characteristics of recent representative localization approaches are discussed and presented, and the directions of research in this area are introduced.
Wireless Sensor Networks, a novel technology about acquiring and processing information, have been proposed for a multitude of diverse applications. The problem of self-localization, that is, determining where a given node is physically or relatively located in the networks, is a challenging one, and yet extremely crucial for many applications. In this paper, the evaluation criterion of the performance and the taxonomy for wireless sensor networks self-localization systems and algorithms are described, the principles and characteristics of recent representative localization approaches are discussed and presented, and the directions of research in this area are introduced.
2013,24(1):77-90
, DOI: 10.3724/SP.J.1001.2013.04339
Abstract:
Task parallel programming model is a widely used parallel programming model on multi-core platforms. With the intention of simplifying parallel programming and improving the utilization of multiple cores, this paper provides an introduction to the essential programming interfaces and the supporting mechanism used in task parallel programming models and discusses issues and the latest achievements from three perspectives: Parallelism expression, data management and task scheduling. In the end, some future trends in this area are discussed.
Task parallel programming model is a widely used parallel programming model on multi-core platforms. With the intention of simplifying parallel programming and improving the utilization of multiple cores, this paper provides an introduction to the essential programming interfaces and the supporting mechanism used in task parallel programming models and discusses issues and the latest achievements from three perspectives: Parallelism expression, data management and task scheduling. In the end, some future trends in this area are discussed.
2011,22(1):115-131
, DOI: 10.3724/SP.J.1001.2011.03950
Abstract:
The Internet traffic model is the key issue for network performance management, Quality of Service management, and admission control. The paper first summarizes the primary characteristics of Internet traffic, as well as the metrics of Internet traffic. It also illustrates the significance and classification of traffic modeling. Next, the paper chronologically categorizes the research activities of traffic modeling into three phases: 1) traditional Poisson modeling; 2) self-similar modeling; and 3) new research debates and new progress. Thorough reviews of the major research achievements of each phase are conducted. Finally, the paper identifies some open research issue and points out possible future research directions in traffic modeling area.
The Internet traffic model is the key issue for network performance management, Quality of Service management, and admission control. The paper first summarizes the primary characteristics of Internet traffic, as well as the metrics of Internet traffic. It also illustrates the significance and classification of traffic modeling. Next, the paper chronologically categorizes the research activities of traffic modeling into three phases: 1) traditional Poisson modeling; 2) self-similar modeling; and 3) new research debates and new progress. Thorough reviews of the major research achievements of each phase are conducted. Finally, the paper identifies some open research issue and points out possible future research directions in traffic modeling area.
2017,28(4):959-992
, DOI: 10.13328/j.cnki.jos.005143
Abstract:
The development of mobile internet and the popularity of mobile terminals produce massive trajectory data of moving objects under the era of big data. Trajectory data has spatio-temporal characteristics and rich information. Trajectory data processing techniques can be used to mine the patterns of human activities and behaviors, the moving patterns of vehicles in the city and the changes of atmospheric environment. However, trajectory data also can be exploited to disclose moving objects' privacy information (e.g., behaviors, hobbies and social relationships). Accordingly, attackers can easily access moving objects' privacy information by digging into their trajectory data such as activities and check-in locations. In another front of research, quantum computation presents an important theoretical direction to mine big data due to its scalable and powerful storage and computing capacity. Applying quantum computing approaches to handle trajectory big data could make some complex problem solvable and achieve higher efficiency. This paper reviews the key technologies of processing trajectory data. First the concept and characteristics of trajectory data is introduced, and the pre-processing methods, including noise filtering and data compression, are summarized. Then, the trajectory indexing and querying techniques, and the current achievements of mining trajectory data, such as pattern mining and trajectory classification, are reviewed. Next, an overview of the basic theories and characteristics of privacy preserving with respect to trajectory data is provided. The supporting techniques of trajectory big data mining, such as processing framework and data visualization, are presented in detail. Some possible ways of applying quantum computation into trajectory data processing, as well as the implementation of some core trajectory mining algorithms by quantum computation are also described. Finally, the challenges of trajectory data processing and promising future research directions are discussed.
The development of mobile internet and the popularity of mobile terminals produce massive trajectory data of moving objects under the era of big data. Trajectory data has spatio-temporal characteristics and rich information. Trajectory data processing techniques can be used to mine the patterns of human activities and behaviors, the moving patterns of vehicles in the city and the changes of atmospheric environment. However, trajectory data also can be exploited to disclose moving objects' privacy information (e.g., behaviors, hobbies and social relationships). Accordingly, attackers can easily access moving objects' privacy information by digging into their trajectory data such as activities and check-in locations. In another front of research, quantum computation presents an important theoretical direction to mine big data due to its scalable and powerful storage and computing capacity. Applying quantum computing approaches to handle trajectory big data could make some complex problem solvable and achieve higher efficiency. This paper reviews the key technologies of processing trajectory data. First the concept and characteristics of trajectory data is introduced, and the pre-processing methods, including noise filtering and data compression, are summarized. Then, the trajectory indexing and querying techniques, and the current achievements of mining trajectory data, such as pattern mining and trajectory classification, are reviewed. Next, an overview of the basic theories and characteristics of privacy preserving with respect to trajectory data is provided. The supporting techniques of trajectory big data mining, such as processing framework and data visualization, are presented in detail. Some possible ways of applying quantum computation into trajectory data processing, as well as the implementation of some core trajectory mining algorithms by quantum computation are also described. Finally, the challenges of trajectory data processing and promising future research directions are discussed.
2010,21(2):344-358
Abstract:
In this paper, the existing intrusion tolerance and self-destruction technology are integrated into autonomic computing in order to construct an autonomic dependability model based on SM-PEPA (semi-Markov performance evaluation process algebra) which is capable of formal analysis and verification. It can hierarchically anticipate Threats to dependability (TtD) at different levels in a self-management manner to satisfy the special requirements for dependability of mission-critical systems. Based on this model, a quantification approach is proposed on the view of steady-state probability to evaluate autonomic dependability. Finally, this paper analyzes the impacts of parameters of the model on autonomic dependability in a case study, and the experimental results demonstrate that improving the detection rate of TtD as well as the successful rate of self-healing will greatly increase the autonomic dependability.
In this paper, the existing intrusion tolerance and self-destruction technology are integrated into autonomic computing in order to construct an autonomic dependability model based on SM-PEPA (semi-Markov performance evaluation process algebra) which is capable of formal analysis and verification. It can hierarchically anticipate Threats to dependability (TtD) at different levels in a self-management manner to satisfy the special requirements for dependability of mission-critical systems. Based on this model, a quantification approach is proposed on the view of steady-state probability to evaluate autonomic dependability. Finally, this paper analyzes the impacts of parameters of the model on autonomic dependability in a case study, and the experimental results demonstrate that improving the detection rate of TtD as well as the successful rate of self-healing will greatly increase the autonomic dependability.
2011,22(6):1299-1315
, DOI: 10.3724/SP.J.1001.2011.03993
Abstract:
Attribute-Based encryption (ABE) scheme takes attributes as the public key and associates the ciphertext and user’s secret key with attributes, so that it can support expressive access control policies. This dramatically reduces the cost of network bandwidth and sending node’s operation in fine-grained access control of data sharing. Therefore, ABE has a broad prospect of application in the area of fine-grained access control. After analyzing the basic ABE system and its two variants, Key-Policy ABE (KP-ABE) and Ciphertext-Policy ABE (CP-ABE), this study elaborates the research problems relating to ABE systems, including access structure design for CP-ABE, attribute key revocation, key abuse and multi-authorities ABE with an extensive comparison of their functionality and performance. Finally, this study discusses the need-to-be solved problems and main research directions in ABE.
Attribute-Based encryption (ABE) scheme takes attributes as the public key and associates the ciphertext and user’s secret key with attributes, so that it can support expressive access control policies. This dramatically reduces the cost of network bandwidth and sending node’s operation in fine-grained access control of data sharing. Therefore, ABE has a broad prospect of application in the area of fine-grained access control. After analyzing the basic ABE system and its two variants, Key-Policy ABE (KP-ABE) and Ciphertext-Policy ABE (CP-ABE), this study elaborates the research problems relating to ABE systems, including access structure design for CP-ABE, attribute key revocation, key abuse and multi-authorities ABE with an extensive comparison of their functionality and performance. Finally, this study discusses the need-to-be solved problems and main research directions in ABE.
2009,20(3):524-545
Abstract:
Nowadays it has been widely accepted that the quality of software highly depends on the process that iscarried out in an organization. As part of the effort to support software process engineering activities, the researchon software process modeling and analysis is to provide an effective means to represent and analyze a process and,by doing so, to enhance the understanding of the modeled process. In addition, an enactable process model canprovide a direct guidance for the actual development process. Thus, the enforcement of the process model candirectly contribute to the improvement of the software quality. In this paper, a systematic review is carried out tosurvey the recent development in software process modeling. 72 papers from 20 conference proceedings and 7journals are identified as the evidence. The review aims to promote a better understanding of the literature byanswering the following three questions: 1) What kinds of paradigms are existing methods based on? 2) What kinds of purposes does the existing research have? 3) What kinds of new trends are reflected in the current research? Afterproviding the systematic review, we present our software process modeling method based on a multi-dimensionaland integration methodology that is intended to address several core issues facing the community.
Nowadays it has been widely accepted that the quality of software highly depends on the process that iscarried out in an organization. As part of the effort to support software process engineering activities, the researchon software process modeling and analysis is to provide an effective means to represent and analyze a process and,by doing so, to enhance the understanding of the modeled process. In addition, an enactable process model canprovide a direct guidance for the actual development process. Thus, the enforcement of the process model candirectly contribute to the improvement of the software quality. In this paper, a systematic review is carried out tosurvey the recent development in software process modeling. 72 papers from 20 conference proceedings and 7journals are identified as the evidence. The review aims to promote a better understanding of the literature byanswering the following three questions: 1) What kinds of paradigms are existing methods based on? 2) What kinds of purposes does the existing research have? 3) What kinds of new trends are reflected in the current research? Afterproviding the systematic review, we present our software process modeling method based on a multi-dimensionaland integration methodology that is intended to address several core issues facing the community.
2014,25(1):37-50
, DOI: 10.13328/j.cnki.jos.004497
Abstract:
This paper surveys the state of the art of speech emotion recognition (SER), and presents an outlook on the trend of future SER technology. First, the survey summarizes and analyzes SER in detail from five perspectives, including emotion representation models, representative emotional speech corpora, emotion-related acoustic features extraction, SER methods and applications. Then, based on the survey, the challenges faced by current SER research are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, and presents detailed comparison and analysis between these methods.
This paper surveys the state of the art of speech emotion recognition (SER), and presents an outlook on the trend of future SER technology. First, the survey summarizes and analyzes SER in detail from five perspectives, including emotion representation models, representative emotional speech corpora, emotion-related acoustic features extraction, SER methods and applications. Then, based on the survey, the challenges faced by current SER research are concluded. This paper aims to take a deep insight into the mainstream methods and recent progress in this field, and presents detailed comparison and analysis between these methods.
2018,29(10):2966-2994
, DOI: 10.13328/j.cnki.jos.005551
Abstract:
In recent years, the rapid development of Internet technology and Web applications has triggered the explosion of various data on the Internet, which generates a large amount of valuable knowledge. How to organize, represent and analyze these knowledge has attracted much attention. Knowledge graph was thus developed to organize these knowledge in a semantical and visualized manner. Knowledge reasoning over knowledge graph then becomes one of the hot research topics and plays an important role in many applications such as vertical search and intelligent question-answer. The goal of knowledge reasoning over knowledge graph is to infer new facts or identify erroneous facts according to existing ones. Unlike traditional knowledge reasoning, knowledge reasoning over knowledge graph is more diversified, due to the simplicity, intuitiveness, flexibility, and richness of knowledge representation in knowledge graph. Starting with the basic concept of knowledge reasoning, this paper presents a survey on the recently developed methods for knowledge reasoning over knowledge graph. Specifically, the research progress is reviewed in detail from two aspects:One-Step reasoning and multi-step reasoning, each including rule based reasoning, distributed embedding based reasoning, neural network based reasoning and hybrid reasoning. Finally, future research directions and outlook of knowledge reasoning over knowledge graph are discussed.
In recent years, the rapid development of Internet technology and Web applications has triggered the explosion of various data on the Internet, which generates a large amount of valuable knowledge. How to organize, represent and analyze these knowledge has attracted much attention. Knowledge graph was thus developed to organize these knowledge in a semantical and visualized manner. Knowledge reasoning over knowledge graph then becomes one of the hot research topics and plays an important role in many applications such as vertical search and intelligent question-answer. The goal of knowledge reasoning over knowledge graph is to infer new facts or identify erroneous facts according to existing ones. Unlike traditional knowledge reasoning, knowledge reasoning over knowledge graph is more diversified, due to the simplicity, intuitiveness, flexibility, and richness of knowledge representation in knowledge graph. Starting with the basic concept of knowledge reasoning, this paper presents a survey on the recently developed methods for knowledge reasoning over knowledge graph. Specifically, the research progress is reviewed in detail from two aspects:One-Step reasoning and multi-step reasoning, each including rule based reasoning, distributed embedding based reasoning, neural network based reasoning and hybrid reasoning. Finally, future research directions and outlook of knowledge reasoning over knowledge graph are discussed.
2013,24(4):825-842
, DOI: 10.3724/SP.J.1001.2013.04369
Abstract:
Honeypot is a proactive defense technology, introduced by the defense side to change the asymmetric situation of a network attack and defensive game. Through the deployment of the honeypots, i.e. security resources without any production purpose, the defenders can deceive attackers to illegally take advantage of the honeypots and capture and analyze the attack behaviors to understand the attack tools and methods, and to learn the intentions and motivations. Honeypot technology has won the sustained attention of the security community to make considerable progress and get wide application, and has become one of the main technical means of the Internet security threat monitoring and analysis. In this paper, the origin and evolution process of the honeypot technology are presented first. Next, the key mechanisms of honeypot technology are comprehensively analyzed, the development process of the honeypot deployment structure is also reviewed, and the latest applications of honeypot technology in the directions of Internet security threat monitoring, analysis and prevention are summarized. Finally, the problems of honeypot technology, development trends and further research directions are discussed.
Honeypot is a proactive defense technology, introduced by the defense side to change the asymmetric situation of a network attack and defensive game. Through the deployment of the honeypots, i.e. security resources without any production purpose, the defenders can deceive attackers to illegally take advantage of the honeypots and capture and analyze the attack behaviors to understand the attack tools and methods, and to learn the intentions and motivations. Honeypot technology has won the sustained attention of the security community to make considerable progress and get wide application, and has become one of the main technical means of the Internet security threat monitoring and analysis. In this paper, the origin and evolution process of the honeypot technology are presented first. Next, the key mechanisms of honeypot technology are comprehensively analyzed, the development process of the honeypot deployment structure is also reviewed, and the latest applications of honeypot technology in the directions of Internet security threat monitoring, analysis and prevention are summarized. Finally, the problems of honeypot technology, development trends and further research directions are discussed.
2018,29(10):3068-3090
, DOI: 10.13328/j.cnki.jos.005607
Abstract:
Designing problems are ubiquitous in science research and industry applications. In recent years, Bayesian optimization, which acts as a very effective global optimization algorithm, has been widely applied in designing problems. By structuring the probabilistic surrogate model and the acquisition function appropriately, Bayesian optimization framework can guarantee to obtain the optimal solution under a few numbers of function evaluations, thus it is very suitable to solve the extremely complex optimization problems in which their objective functions could not be expressed, or the functions are non-convex, multimodal and computational expensive. This paper provides a detailed analysis on Bayesian optimization in methodology and application areas, and discusses its research status and the problems in future researches. This work is hopefully beneficial to the researchers from the related communities.
Designing problems are ubiquitous in science research and industry applications. In recent years, Bayesian optimization, which acts as a very effective global optimization algorithm, has been widely applied in designing problems. By structuring the probabilistic surrogate model and the acquisition function appropriately, Bayesian optimization framework can guarantee to obtain the optimal solution under a few numbers of function evaluations, thus it is very suitable to solve the extremely complex optimization problems in which their objective functions could not be expressed, or the functions are non-convex, multimodal and computational expensive. This paper provides a detailed analysis on Bayesian optimization in methodology and application areas, and discusses its research status and the problems in future researches. This work is hopefully beneficial to the researchers from the related communities.
2004,15(11):1583-1594
Abstract:
Uncertainty exists widely in the subjective and objective world. In all kinds of uncertainty, randomness and fuzziness are the most important and fundamental. In this paper, the relationship between randomness and fuzziness is discussed. Uncertain states and their changes can be measured by entropy and hyper-entropy respectively. Taken advantage of entropy and hyper-entropy, the uncertainty of chaos, fractal and complex networks by their various evolution and differentiation are further studied. A simple and effective way is proposed to simulate the uncertainty by means of knowledge representation which provides a basis for the automation of both logic and image thinking with uncertainty. The AI (artificial intelligence) with uncertainty is a new cross-discipline, which covers computer science, physics, mathematics, brain science, psychology, cognitive science, biology and philosophy, and results in the automation of representation, process and thinking for uncertain information and knowledge.
Uncertainty exists widely in the subjective and objective world. In all kinds of uncertainty, randomness and fuzziness are the most important and fundamental. In this paper, the relationship between randomness and fuzziness is discussed. Uncertain states and their changes can be measured by entropy and hyper-entropy respectively. Taken advantage of entropy and hyper-entropy, the uncertainty of chaos, fractal and complex networks by their various evolution and differentiation are further studied. A simple and effective way is proposed to simulate the uncertainty by means of knowledge representation which provides a basis for the automation of both logic and image thinking with uncertainty. The AI (artificial intelligence) with uncertainty is a new cross-discipline, which covers computer science, physics, mathematics, brain science, psychology, cognitive science, biology and philosophy, and results in the automation of representation, process and thinking for uncertain information and knowledge.
2019,30(2):440-468
, DOI: 10.13328/j.cnki.jos.005659
Abstract:
Recent years, applying Deep Learning (DL) into Image Semantic Segmentation (ISS) has been widely used due to its state-of-the-art performances and high-quality results. This paper systematically reviews the contribution of DL to the field of ISS. Different methods of ISS based on DL (ISSbDL) are summarized. These methods are divided into ISS based on the Regional Classification (ISSbRC) and ISS based on the Pixel Classification (ISSbPC) according to the image segmentation characteristics and segmentation granularity. Then, the methods of ISSbPC are surveyed from two points of view:ISS based on Fully Supervised Learning (ISSbFSL) and ISS based on Weakly Supervised Learning (ISSbWSL). The representative algorithms of each method are introduced and analyzed, as well as the basic workflow, framework, advantages and disadvantages of these methods are detailedly analyzed and compared. In addition, the related experiments of ISS are analyzed and summarized, and the common data sets and performance evaluation indexes in ISS experiments are introduced. Finally, possible research directions and trends are given and analyzed.
Recent years, applying Deep Learning (DL) into Image Semantic Segmentation (ISS) has been widely used due to its state-of-the-art performances and high-quality results. This paper systematically reviews the contribution of DL to the field of ISS. Different methods of ISS based on DL (ISSbDL) are summarized. These methods are divided into ISS based on the Regional Classification (ISSbRC) and ISS based on the Pixel Classification (ISSbPC) according to the image segmentation characteristics and segmentation granularity. Then, the methods of ISSbPC are surveyed from two points of view:ISS based on Fully Supervised Learning (ISSbFSL) and ISS based on Weakly Supervised Learning (ISSbWSL). The representative algorithms of each method are introduced and analyzed, as well as the basic workflow, framework, advantages and disadvantages of these methods are detailedly analyzed and compared. In addition, the related experiments of ISS are analyzed and summarized, and the common data sets and performance evaluation indexes in ISS experiments are introduced. Finally, possible research directions and trends are given and analyzed.
2011,22(3):381-407
, DOI: 10.3724/SP.J.1001.2011.03934
Abstract:
The popularity of the Internet and the boom of the World Wide Web foster innovative changes in software technology that give birth to a new form of software—networked software, which delivers diversified and personalized on-demand services to the public. With the ever-increasing expansion of applications and users, the scale and complexity of networked software are growing beyond the information processing capability of human beings, which brings software engineers a series of challenges to face. In order to come to a scientific understanding of this kind of ultra-large-scale artificial complex systems, a survey research on the infrastructure, application services, and social interactions of networked software is conducted from a three-dimensional perspective of cyberization, servicesation, and socialization. Interestingly enough, most of them have been found to share the same global characteristics of complex networks such as “Small World” and “Scale Free”. Next, the impact of the empirical study on software engineering research and practice and its implications for further investigations are systematically set forth. The convergence of software engineering and other disciplines will put forth new ideas and thoughts that will breed a new way of thinking and input new methodologies for the study of networked software. This convergence is also expected to achieve the innovations of theories, methods, and key technologies of software engineering to promote the rapid development of software service industry in China.
The popularity of the Internet and the boom of the World Wide Web foster innovative changes in software technology that give birth to a new form of software—networked software, which delivers diversified and personalized on-demand services to the public. With the ever-increasing expansion of applications and users, the scale and complexity of networked software are growing beyond the information processing capability of human beings, which brings software engineers a series of challenges to face. In order to come to a scientific understanding of this kind of ultra-large-scale artificial complex systems, a survey research on the infrastructure, application services, and social interactions of networked software is conducted from a three-dimensional perspective of cyberization, servicesation, and socialization. Interestingly enough, most of them have been found to share the same global characteristics of complex networks such as “Small World” and “Scale Free”. Next, the impact of the empirical study on software engineering research and practice and its implications for further investigations are systematically set forth. The convergence of software engineering and other disciplines will put forth new ideas and thoughts that will breed a new way of thinking and input new methodologies for the study of networked software. This convergence is also expected to achieve the innovations of theories, methods, and key technologies of software engineering to promote the rapid development of software service industry in China.
2010,21(7):1605-1619
Abstract:
The rapid development of Internet leads to an increase in system complexity and uncertainty. Traditional network management can not meet the requirement, and it shall evolve to fusion based Cyberspace Situational Awareness (CSA). Based on the analysis of function shortage and development requirement, this paper introduces CSA as well as its origin, conception, objective and characteristics. Firstly, a CSA research framework is proposed and the research history is investigated, based on which the main aspects and the existing issues of the research are analyzed. Meanwhile, assessment methods are divided into three categories: Mathematics model, knowledge reasoning and pattern recognition. Then, this paper discusses CSA from three aspects: Model, knowledge representation and assessment methods, and then goes into detail about main idea, assessment process, merits and shortcomings of novel methods. Many typical methods are compared. The current application research of CSA in the fields of security, transmission, survivable, system evaluation and so on is presented. Finally, this paper points the development directions of CSA and offers the conclusions from issue system, technical system and application system.
The rapid development of Internet leads to an increase in system complexity and uncertainty. Traditional network management can not meet the requirement, and it shall evolve to fusion based Cyberspace Situational Awareness (CSA). Based on the analysis of function shortage and development requirement, this paper introduces CSA as well as its origin, conception, objective and characteristics. Firstly, a CSA research framework is proposed and the research history is investigated, based on which the main aspects and the existing issues of the research are analyzed. Meanwhile, assessment methods are divided into three categories: Mathematics model, knowledge reasoning and pattern recognition. Then, this paper discusses CSA from three aspects: Model, knowledge representation and assessment methods, and then goes into detail about main idea, assessment process, merits and shortcomings of novel methods. Many typical methods are compared. The current application research of CSA in the fields of security, transmission, survivable, system evaluation and so on is presented. Finally, this paper points the development directions of CSA and offers the conclusions from issue system, technical system and application system.
2018,29(7):2092-2115
, DOI: 10.13328/j.cnki.jos.005589
Abstract:
Blockchain is a distributed public ledger technology that originates from the digital cryptocurrency, bitcoin. Its development has attracted wide attention in industry and academia fields. Blockchain has the advantages of de-centralization, trustworthiness, anonymity and immutability. It breaks through the limitation of traditional center-based technology and has broad development prospect. This paper introduces the research progress of blockchain technology and its application in the field of information security. Firstly, the basic theory and model of blockchain are introduced from five aspects:Basic framework, key technology, technical feature, and application mode and area. Secondly, from the perspective of current research situation of blockchain in the field of information security, this paper summarizes the research progress of blockchain in authentication technology, access control technology and data protection technology, and compares the characteristics of various researches. Finally, the application challenges of blockchain technology are analyzed, and the development outlook of blockchain in the field of information security is highlighted. This study intends to provide certain reference value for future research work.
Blockchain is a distributed public ledger technology that originates from the digital cryptocurrency, bitcoin. Its development has attracted wide attention in industry and academia fields. Blockchain has the advantages of de-centralization, trustworthiness, anonymity and immutability. It breaks through the limitation of traditional center-based technology and has broad development prospect. This paper introduces the research progress of blockchain technology and its application in the field of information security. Firstly, the basic theory and model of blockchain are introduced from five aspects:Basic framework, key technology, technical feature, and application mode and area. Secondly, from the perspective of current research situation of blockchain in the field of information security, this paper summarizes the research progress of blockchain in authentication technology, access control technology and data protection technology, and compares the characteristics of various researches. Finally, the application challenges of blockchain technology are analyzed, and the development outlook of blockchain in the field of information security is highlighted. This study intends to provide certain reference value for future research work.
2009,20(1):124-137
Abstract:
The appearance of plenty of intelligent devices equipped for short-range wireless communications boosts the fast rise of wireless ad hoc networks application. However, in many realistic application environments, nodes form a disconnected network for most of the time due to nodal mobility, low density, lossy link, etc. Conventional communication model of mobile ad hoc network (MANET) requires at least one path existing from source to destination nodes, which results in communication failure in these scenarios. Opportunistic networks utilize the communication opportunities arising from node movement to forward messages in a hop-by-hop way, and implement communications between nodes based on the "store-carry-forward" routing pattern. This networking approach, totally different from the traditional communication model, captures great interests from researchers. This paper first introduces the conceptions and theories of opportunistic networks and some current typical applications. Then it elaborates the popular research problems including opportunistic forwarding mechanism, mobility model and opportunistic data dissemination and retrieval. Some other interesting research points such as communication middleware, cooperation and security problem and new applications are stated briefly. Finally, the paper concludes and looks forward to the possible research focuses for opportunistic networks in the future.
The appearance of plenty of intelligent devices equipped for short-range wireless communications boosts the fast rise of wireless ad hoc networks application. However, in many realistic application environments, nodes form a disconnected network for most of the time due to nodal mobility, low density, lossy link, etc. Conventional communication model of mobile ad hoc network (MANET) requires at least one path existing from source to destination nodes, which results in communication failure in these scenarios. Opportunistic networks utilize the communication opportunities arising from node movement to forward messages in a hop-by-hop way, and implement communications between nodes based on the "store-carry-forward" routing pattern. This networking approach, totally different from the traditional communication model, captures great interests from researchers. This paper first introduces the conceptions and theories of opportunistic networks and some current typical applications. Then it elaborates the popular research problems including opportunistic forwarding mechanism, mobility model and opportunistic data dissemination and retrieval. Some other interesting research points such as communication middleware, cooperation and security problem and new applications are stated briefly. Finally, the paper concludes and looks forward to the possible research focuses for opportunistic networks in the future.
2005,16(1):1-7
Abstract:
The paper gives some thinking according to the following four aspects: 1) from the law of things development, revealing the development history of software engineering technology; 2) from the point of software natural characteristic, analyzing the construction of every abstraction layer of virtual machine; 3) from the point of software development, proposing the research content of software engineering discipline, and research the pattern of industrialized software production; 4) based on the appearance of Internet technology, exploring the development trend of software technology.
The paper gives some thinking according to the following four aspects: 1) from the law of things development, revealing the development history of software engineering technology; 2) from the point of software natural characteristic, analyzing the construction of every abstraction layer of virtual machine; 3) from the point of software development, proposing the research content of software engineering discipline, and research the pattern of industrialized software production; 4) based on the appearance of Internet technology, exploring the development trend of software technology.
2014,25(4):839-862
, DOI: 10.13328/j.cnki.jos.004558
Abstract:
Batch computing and stream computing are two important forms of big data computing. The research and discussions on batch computing in big data environment are comparatively sufficient. But how to efficiently deal with stream computing to meet many requirements, such as low latency, high throughput and continuously reliable running, and how to build efficient stream big data computing systems, are great challenges in the big data computing research. This paper provides a research of the data computing architecture and the key issues in stream computing in big data environments. Firstly, the research gives a brief summary of three application scenarios of stream computing in business intelligence, marketing and public service. It also shows distinctive features of the stream computing in big data environment, such as real time, volatility, burstiness, irregularity and infinity. A well-designed stream computing system always optimizes in system structure, data transmission, application interfaces, high-availability, and so on. Subsequently, the research offers detailed analyses and comparisons of five typical and open-source stream computing systems in big data environment. Finally, the research specifically addresses some new challenges of the stream big data systems, such as scalability, fault tolerance, consistency, load balancing and throughput.
Batch computing and stream computing are two important forms of big data computing. The research and discussions on batch computing in big data environment are comparatively sufficient. But how to efficiently deal with stream computing to meet many requirements, such as low latency, high throughput and continuously reliable running, and how to build efficient stream big data computing systems, are great challenges in the big data computing research. This paper provides a research of the data computing architecture and the key issues in stream computing in big data environments. Firstly, the research gives a brief summary of three application scenarios of stream computing in business intelligence, marketing and public service. It also shows distinctive features of the stream computing in big data environment, such as real time, volatility, burstiness, irregularity and infinity. A well-designed stream computing system always optimizes in system structure, data transmission, application interfaces, high-availability, and so on. Subsequently, the research offers detailed analyses and comparisons of five typical and open-source stream computing systems in big data environment. Finally, the research specifically addresses some new challenges of the stream big data systems, such as scalability, fault tolerance, consistency, load balancing and throughput.
2013,24(8):1786-1803
, DOI: 10.3724/SP.J.1001.2013.04416
Abstract:
Many specific application oriented NoSQL database systems are developed for satisfying the new requirement of big data management. This paper surveys researches on typical NoSQL database based on key-value data model. First, the characteristics of big data, and the key technique issues supporting big data management are introduced. Then frontier efforts and research challenges are given, including system architecture, data model, access mode, index, transaction, system elasticity, load balance, replica strategy, data consistency, flash cache, MapReduce based data process and new generation data management system etc. Finally, research prospects are given.
Many specific application oriented NoSQL database systems are developed for satisfying the new requirement of big data management. This paper surveys researches on typical NoSQL database based on key-value data model. First, the characteristics of big data, and the key technique issues supporting big data management are introduced. Then frontier efforts and research challenges are given, including system architecture, data model, access mode, index, transaction, system elasticity, load balance, replica strategy, data consistency, flash cache, MapReduce based data process and new generation data management system etc. Finally, research prospects are given.
2016,27(11):2855-2869
, DOI: 10.13328/j.cnki.jos.004932
Abstract:
With the proliferation of the Chinese social network (especially the rise of weibo), the productivity and lifestyle of the country's society is more and more profoundly influenced by the Chinese internet public events. Due to the lack of the effective technical means, the efficiency of information processing is limited. This paper proposes a public event information entropy calculation method. First, a mathematical modeling of event information content is built. Then, multidimensional random variable information entropy of the public events is calculated based on Shannon information theory. Furthermore, a new technical index of quantitative analysis to the internet public events is put forward, laying out a foundation for further research work.
With the proliferation of the Chinese social network (especially the rise of weibo), the productivity and lifestyle of the country's society is more and more profoundly influenced by the Chinese internet public events. Due to the lack of the effective technical means, the efficiency of information processing is limited. This paper proposes a public event information entropy calculation method. First, a mathematical modeling of event information content is built. Then, multidimensional random variable information entropy of the public events is calculated based on Shannon information theory. Furthermore, a new technical index of quantitative analysis to the internet public events is put forward, laying out a foundation for further research work.
2006,17(9):1848-1859
Abstract:
In recent years, there have been extensive studies and rapid progresses in automatic text categorization, which is one of the hotspots and key techniques in the information retrieval and data mining field. Highlighting the state-of-art challenging issues and research trends for content information processing of Internet and other complex applications, this paper presents a survey on the up-to-date development in text categorization based on machine learning, including model, algorithm and evaluation. It is pointed out that problems such as nonlinearity, skewed data distribution, labeling bottleneck, hierarchical categorization, scalability of algorithms and categorization of Web pages are the key problems to the study of text categorization. Possible solutions to these problems are also discussed respectively. Finally, some future directions of research are given.
In recent years, there have been extensive studies and rapid progresses in automatic text categorization, which is one of the hotspots and key techniques in the information retrieval and data mining field. Highlighting the state-of-art challenging issues and research trends for content information processing of Internet and other complex applications, this paper presents a survey on the up-to-date development in text categorization based on machine learning, including model, algorithm and evaluation. It is pointed out that problems such as nonlinearity, skewed data distribution, labeling bottleneck, hierarchical categorization, scalability of algorithms and categorization of Web pages are the key problems to the study of text categorization. Possible solutions to these problems are also discussed respectively. Finally, some future directions of research are given.
2023,34(2):625-654
, DOI: 10.13328/j.cnki.jos.006696
Abstract:
Source code bug (vulnerability) detection is a process of judging whether there are unexpected behaviors in the program code. It is widely used in software engineering tasks such as software testing and software maintenance, and plays a vital role in software functional assurance and application security. Traditional vulnerability detection research is based on program analysis, which usually requires strong domain knowledge and complex calculation rules, and faces the problem of state explosion, resulting in limited detection performance, and there is room for greater improvement in the rate of false positives and false negatives. In recent years, the open source community's vigorous development has accumulated massive amounts of data with open source code as the core. In this context, the feature learning capabilities of deep learning can automatically learn semantically rich code representations, thereby providing a new way for vulnerability detection. This study collected the latest high-level papers in this field, systematically summarized and explained the current methods from two aspects:vulnerability code dataset and deep learning vulnerability detection model. Finally, it summarizes the main challenges faced by the research in this field, and looks forward to the possible future research focus.
Source code bug (vulnerability) detection is a process of judging whether there are unexpected behaviors in the program code. It is widely used in software engineering tasks such as software testing and software maintenance, and plays a vital role in software functional assurance and application security. Traditional vulnerability detection research is based on program analysis, which usually requires strong domain knowledge and complex calculation rules, and faces the problem of state explosion, resulting in limited detection performance, and there is room for greater improvement in the rate of false positives and false negatives. In recent years, the open source community's vigorous development has accumulated massive amounts of data with open source code as the core. In this context, the feature learning capabilities of deep learning can automatically learn semantically rich code representations, thereby providing a new way for vulnerability detection. This study collected the latest high-level papers in this field, systematically summarized and explained the current methods from two aspects:vulnerability code dataset and deep learning vulnerability detection model. Finally, it summarizes the main challenges faced by the research in this field, and looks forward to the possible future research focus.
2020,31(7):2127-2156
, DOI: 10.13328/j.cnki.jos.006052
Abstract:
Machine learning has become a core technology in areas such as big data, Internet of Things, and cloud computing. Training machine learning models requires a large amount of data, which is often collected by means of crowdsourcing and contains a large number of private data including personally identifiable information (such as phone number, id number, etc.) and sensitive information (such as financial data, health care, etc.). How to protect these data with low cost and high efficiency is an important issue. This paper first introduces the concept of machine learning, explains various definitions of privacy in machine learning and demonstrates all kinds of privacy threats encountered in machine learning, then continues to elaborate on the working principle and outstanding features of the mainstream technology of machine learning privacy protection. According to differential privacy, homomorphic encryption, and secure multi-party computing, the research achievements in the field of machine learning privacy protection are summarized respectively. On this basis, the paper comparatively analyzes the main advantages and disadvantages of different mechanisms of privacy preserving for machine learning. Finally, the developing trend of privacy preserving for machine learning is prospected, and the possible research directions in this field are proposed.
Machine learning has become a core technology in areas such as big data, Internet of Things, and cloud computing. Training machine learning models requires a large amount of data, which is often collected by means of crowdsourcing and contains a large number of private data including personally identifiable information (such as phone number, id number, etc.) and sensitive information (such as financial data, health care, etc.). How to protect these data with low cost and high efficiency is an important issue. This paper first introduces the concept of machine learning, explains various definitions of privacy in machine learning and demonstrates all kinds of privacy threats encountered in machine learning, then continues to elaborate on the working principle and outstanding features of the mainstream technology of machine learning privacy protection. According to differential privacy, homomorphic encryption, and secure multi-party computing, the research achievements in the field of machine learning privacy protection are summarized respectively. On this basis, the paper comparatively analyzes the main advantages and disadvantages of different mechanisms of privacy preserving for machine learning. Finally, the developing trend of privacy preserving for machine learning is prospected, and the possible research directions in this field are proposed.
2005,16(10):1743-1756
Abstract:
This paper presents a survey on the theory of provable security and its applications to the design and analysis of security protocols. It clarifies what the provable security is, explains some basic notions involved in the theory of provable security and illustrates the basic idea of random oracle model. It also reviews the development and advances of provably secure public-key encryption and digital signature schemes, in the random oracle model or the standard model, as well as the applications of provable security to the design and analysis of session-key distribution protocols and their advances.
This paper presents a survey on the theory of provable security and its applications to the design and analysis of security protocols. It clarifies what the provable security is, explains some basic notions involved in the theory of provable security and illustrates the basic idea of random oracle model. It also reviews the development and advances of provably secure public-key encryption and digital signature schemes, in the random oracle model or the standard model, as well as the applications of provable security to the design and analysis of session-key distribution protocols and their advances.
2013,24(2):295-316
, DOI: 10.3724/SP.J.1001.2013.04336
Abstract:
Under the new application mode, the traditional hierarchy data centers face several limitations in size, bandwidth, scalability, and cost. In order to meet the needs of new applications, data center network should fulfill the requirements with low-cost, such as high scalability, low configuration overhead, robustness and energy-saving. First, the shortcomings of the traditional data center network architecture are summarized, and new requirements are pointed out. Secondly, the existing proposals are divided into two categories, i.e. server-centric and network-centric. Then, several representative architectures of these two categories are overviewed and compared in detail. Finally, the future directions of data center network are discussed.
Under the new application mode, the traditional hierarchy data centers face several limitations in size, bandwidth, scalability, and cost. In order to meet the needs of new applications, data center network should fulfill the requirements with low-cost, such as high scalability, low configuration overhead, robustness and energy-saving. First, the shortcomings of the traditional data center network architecture are summarized, and new requirements are pointed out. Secondly, the existing proposals are divided into two categories, i.e. server-centric and network-centric. Then, several representative architectures of these two categories are overviewed and compared in detail. Finally, the future directions of data center network are discussed.
2012,23(8):2058-2072
, DOI: 10.3724/SP.J.1001.2012.04237
Abstract:
The Distributed denial of service (DDoS) attack is a major threat to the current network. Based on the attack packet level, the study divides DDoS attacks into network-level DDoS attacks and application-level DDoS attacks. Next, the study analyzes the detection and control methods of these two kinds of DDoS attacks in detail, and it also analyzes the drawbacks of different control methods implemented in different network positions. Finally, the study analyzes the drawbacks of the current detection and control methods, the development trend of the DDoS filter system, and corresponding technological challenges are also proposed.
The Distributed denial of service (DDoS) attack is a major threat to the current network. Based on the attack packet level, the study divides DDoS attacks into network-level DDoS attacks and application-level DDoS attacks. Next, the study analyzes the detection and control methods of these two kinds of DDoS attacks in detail, and it also analyzes the drawbacks of different control methods implemented in different network positions. Finally, the study analyzes the drawbacks of the current detection and control methods, the development trend of the DDoS filter system, and corresponding technological challenges are also proposed.
LI Xu-Rong, JI Shou-Ling, WU Chun-Ming, LIU Zhen-Guang, DENG Shui-Guang, CHENG Peng, YANG Min, KONG Xiang-Wei
2021,32(2):496-518
, DOI: 10.13328/j.cnki.jos.006140
Abstract:
Deep learning has achieved great success in the field of computer vision, surpassing many traditional methods. However, in recent years, deep learning technology has been abused in the production of fake videos, making fake videos represented by Deepfakes flooding on the Internet. This technique produces pornographic movies, fake news, political rumors by tampering or replacing the face information of the original videos and synthesizes fake speech. In order to eliminate the negative effects brought by such forgery technologies, many researchers have conducted in-depth research on the identification of fake videos and proposed a series of detection methods to help institutions or communities to identify such fake videos. Nevertheless, the current detection technology still has many limitations such as specific distribution data, specific compression ratio, and so on, far behind the generation technology of fake video. In addition, different researchers handle the problem from different angles. The data sets and evaluation indicators used are not uniform. So far, the academic community still lacks a unified understanding of deep forgery and detection technology. The architecture of deep forgery and detection technology research is not clear. In this review, the development of deep forgery and detection technologies are reviewed. Besides, existing research works are systematically summarize and scientifically classified. Finally, the social risks posed by the spread of Deepfakes technology are discussed, the limitations of detection technology are analyzed, and the challenges and potential research directions of detection technology are discussed, aiming to provide guidance for follow-up researchers to further promote the development and deployment of Deepfakes detection technology.
Deep learning has achieved great success in the field of computer vision, surpassing many traditional methods. However, in recent years, deep learning technology has been abused in the production of fake videos, making fake videos represented by Deepfakes flooding on the Internet. This technique produces pornographic movies, fake news, political rumors by tampering or replacing the face information of the original videos and synthesizes fake speech. In order to eliminate the negative effects brought by such forgery technologies, many researchers have conducted in-depth research on the identification of fake videos and proposed a series of detection methods to help institutions or communities to identify such fake videos. Nevertheless, the current detection technology still has many limitations such as specific distribution data, specific compression ratio, and so on, far behind the generation technology of fake video. In addition, different researchers handle the problem from different angles. The data sets and evaluation indicators used are not uniform. So far, the academic community still lacks a unified understanding of deep forgery and detection technology. The architecture of deep forgery and detection technology research is not clear. In this review, the development of deep forgery and detection technologies are reviewed. Besides, existing research works are systematically summarize and scientifically classified. Finally, the social risks posed by the spread of Deepfakes technology are discussed, the limitations of detection technology are analyzed, and the challenges and potential research directions of detection technology are discussed, aiming to provide guidance for follow-up researchers to further promote the development and deployment of Deepfakes detection technology.
2003,14(9):1621-1628
Abstract:
Recommendation system is one of the most important technologies in E-commerce. With the development of E-commerce, the magnitudes of users and commodities grow rapidly, resulted in the extreme sparsity of user rating data. Traditional similarity measure methods work poor in this situation, make the quality of recommendation system decreased dramatically. To address this issue a novel collaborative filtering algorithm based on item rating prediction is proposed. This method predicts item ratings that users have not rated by the similarity of items, then uses a new similarity measure to find the target users?neighbors. The experimental results show that this method can efficiently improve the extreme sparsity of user rating data, and provid better recommendation results than traditional collaborative filtering algorithms.
Recommendation system is one of the most important technologies in E-commerce. With the development of E-commerce, the magnitudes of users and commodities grow rapidly, resulted in the extreme sparsity of user rating data. Traditional similarity measure methods work poor in this situation, make the quality of recommendation system decreased dramatically. To address this issue a novel collaborative filtering algorithm based on item rating prediction is proposed. This method predicts item ratings that users have not rated by the similarity of items, then uses a new similarity measure to find the target users?neighbors. The experimental results show that this method can efficiently improve the extreme sparsity of user rating data, and provid better recommendation results than traditional collaborative filtering algorithms.
2026,37(5):2167-2201
, DOI: 10.13328/j.cnki.jos.007580
, CSTR: 32375.14.jos.007580
Abstract:
2026,37(5):1982-2005
, DOI: 10.13328/j.cnki.jos.007539
, CSTR: 32375.14.jos.007539
Abstract:
2026,37(4):1591-1614
, DOI: 10.13328/j.cnki.jos.007530
, CSTR: 32375.14.jos.007530
Abstract:
2026,37(4):1854-1884
, DOI: 10.13328/j.cnki.jos.007511
, CSTR: 32375.14.jos.007511
Abstract:
2026,37(3):1316-1339
, DOI: 10.13328/j.cnki.jos.007567
, CSTR: 32375.14.jos.007567
Abstract:
2026,37(3):1290-1315
, DOI: 10.13328/j.cnki.jos.007561
, CSTR: 32375.14.jos.007561
Abstract:
2026,37(3):971-1005
, DOI: 10.13328/j.cnki.jos.007516
, CSTR: 32375.14.jos.007516
Abstract:
2026,37(2):508-542
, DOI: 10.13328/j.cnki.jos.007560
, CSTR: 32375.14.jos.007560
Abstract:
2026,37(2):817-859
, DOI: 10.13328/j.cnki.jos.007506
, CSTR: 32375.14.jos.007506
Abstract:
2026,37(1):157-179
, DOI: 10.13328/j.cnki.jos.007486
, CSTR: 32375.14.jos.007486
Abstract:
2026,37(1):139-156
, DOI: 10.13328/j.cnki.jos.007480
, CSTR: 32375.14.jos.007480
Abstract:
2026,37(1):301-325
, DOI: 10.13328/j.cnki.jos.007478
, CSTR: 32375.14.jos.007478
Abstract:
2026,37(1):464-484
, DOI: 10.13328/j.cnki.jos.007459
, CSTR: 32375.14.jos.007459
Abstract:
2026,37(1):442-463
, DOI: 10.13328/j.cnki.jos.007454
, CSTR: 32375.14.jos.007454
Abstract:
2026,37(1):102-138
, DOI: 10.13328/j.cnki.jos.007452
, CSTR: 32375.14.jos.007452
Abstract:
2026,37(1):62-101
, DOI: 10.13328/j.cnki.jos.007450
, CSTR: 32375.14.jos.007450
Abstract:
2026,37(1):279-300
, DOI: 10.13328/j.cnki.jos.007449
, CSTR: 32375.14.jos.007449
Abstract:
2026,37(1):34-61
, DOI: 10.13328/j.cnki.jos.007446
, CSTR: 32375.14.jos.007446
Abstract:
2026,37(1):425-441
, DOI: 10.13328/j.cnki.jos.007444
, CSTR: 32375.14.jos.007444
Abstract:
2026,37(1):259-278
, DOI: 10.13328/j.cnki.jos.007442
, CSTR: 32375.14.jos.007442
Abstract:
2026,37(1):1-33
, DOI: 10.13328/j.cnki.jos.007441
, CSTR: 32375.14.jos.007441
Abstract:
2026,37(1):200-229
, DOI: 10.13328/j.cnki.jos.007438
, CSTR: 32375.14.jos.007438
Abstract:
2026,37(1):398-424
, DOI: 10.13328/j.cnki.jos.007437
, CSTR: 32375.14.jos.007437
Abstract:
2026,37(1):378-397
, DOI: 10.13328/j.cnki.jos.007422
, CSTR: 32375.14.jos.007422
Abstract:
2026,37(1):230-258
, DOI: 10.13328/j.cnki.jos.007418
, CSTR: 32375.14.jos.007418
Abstract:
2026,37(1):180-199
, DOI: 10.13328/j.cnki.jos.007417
, CSTR: 32375.14.jos.007417
Abstract:
2026,37(1):344-377
, DOI: 10.13328/j.cnki.jos.007413
, CSTR: 32375.14.jos.007413
Abstract:
2026,37(1):326-343
, DOI: 10.13328/j.cnki.jos.007410
, CSTR: 32375.14.jos.007410
Abstract:
2025,36(11):5274-5295
, DOI: 10.13328/j.cnki.jos.007385
, CSTR: 32375.14.jos.007385
Abstract:
2025,36(11):5241-5273
, DOI: 10.13328/j.cnki.jos.007378
, CSTR: 32375.14.jos.007378
Abstract:
2025,36(11):4906-4952
, DOI: 10.13328/j.cnki.jos.007376
, CSTR: 32375.14.jos.007376
Abstract:
2025,36(8):3604-3636
, DOI: 10.13328/j.cnki.jos.007353
, CSTR: 32375.14.jos.007353
Abstract:
2025,36(7):3151-3183
, DOI: 10.13328/j.cnki.jos.007366
, CSTR: 32375.14.jos.007366
Abstract:
2025,36(7):3271-3305
, DOI: 10.13328/j.cnki.jos.007364
, CSTR: 32375.14.jos.007364
Abstract:
2025,36(7):2964-3002
, DOI: 10.13328/j.cnki.jos.007334
, CSTR: 32375.14.jos.007334
Abstract:
2025,36(6):2453-2476
, DOI: 10.13328/j.cnki.jos.007328
, CSTR: 32375.14.jos.007328
Abstract:
2025,36(6):2404-2431
, DOI: 10.13328/j.cnki.jos.007323
, CSTR: 32375.14.jos.007323
Abstract:
2025,36(6):2794-2826
, DOI: 10.13328/j.cnki.jos.007319
, CSTR: 32375.14.jos.007319
Abstract:
2025,36(6):2844-2874
, DOI: 10.13328/j.cnki.jos.007317
, CSTR: 32375.14.jos.007317
Abstract:
2025,36(6):2713-2746
, DOI: 10.13328/j.cnki.jos.007313
, CSTR: 32375.14.jos.007313
Abstract:
2025,36(6):2683-2712
, DOI: 10.13328/j.cnki.jos.007312
, CSTR: 32375.14.jos.007312
Abstract:
2025,36(6):2604-2642
, DOI: 10.13328/j.cnki.jos.007308
, CSTR: 32375.14.jos.007308
Abstract:
2025,36(6):2875-2899
, DOI: 10.13328/j.cnki.jos.007305
, CSTR: 32375.14.jos.007305
Abstract:
2025,36(6):2576-2603
, DOI: 10.13328/j.cnki.jos.007304
, CSTR: 32375.14.jos.007304
Abstract:
2025,36(6):2643-2682
, DOI: 10.13328/j.cnki.jos.007290
, CSTR: 32375.14.jos.007290
Abstract:
2025,36(5):2229-2253
, DOI: 10.13328/j.cnki.jos.007181
, CSTR: 32375.14.jos.007181
Abstract:
2025,36(4):1758-1795
, DOI: 10.13328/j.cnki.jos.007298
, CSTR: 32375.14.jos.007298
Abstract:
2025,36(4):1530-1556
, DOI: 10.13328/j.cnki.jos.007288
, CSTR: 32375.14.jos.007288
Abstract:
2025,36(4):1811-1843
, DOI: 10.13328/j.cnki.jos.007275
, CSTR: 32375.14.jos.007275
Abstract:
2025,36(4):1881-1905
, DOI: 10.13328/j.cnki.jos.007273
, CSTR: 32375.14.jos.007273
Abstract:
2025,36(4):1859-1880
, DOI: 10.13328/j.cnki.jos.007272
, CSTR: 32375.14.jos.007272
Abstract:
2025,36(4):1489-1529
, DOI: 10.13328/j.cnki.jos.007268
, CSTR: 32375.14.jos.007268
Abstract:
2025,36(4):1461-1488
, DOI: 10.13328/j.cnki.jos.007254
, CSTR: 32375.14.jos.007254
Abstract:
2025,36(4):1715-1757
, DOI: 10.13328/j.cnki.jos.007250
, CSTR: 32375.14.jos.007250
Abstract:
2025,36(4):1692-1714
, DOI: 10.13328/j.cnki.jos.007248
, CSTR: 32375.14.jos.007248
Abstract:
2025,36(4):1665-1691
, DOI: 10.13328/j.cnki.jos.007245
, CSTR: 32375.14.jos.007245
Abstract:
2025,36(4):1637-1664
, DOI: 10.13328/j.cnki.jos.007237
, CSTR: 32375.14.jos.007237
Abstract:
2025,36(4):1570-1589
, DOI: 10.13328/j.cnki.jos.007202
, CSTR: 32375.14.jos.007202
Abstract:
2025,36(3):1040-1064
, DOI: 10.13328/j.cnki.jos.007276
, CSTR: 32375.14.jos.007276
Abstract:
2025,36(3):1390-1412
, DOI: 10.13328/j.cnki.jos.007263
, CSTR: 32375.14.jos.007263
Abstract:
2025,36(3):1152-1185
, DOI: 10.13328/j.cnki.jos.007242
, CSTR: 32375.14.jos.007242
Abstract:
2025,36(2):886-914
, DOI: 10.13328/j.cnki.jos.007168
, CSTR: 32375.14.jos.007168
Abstract:
2025,36(1):446-468
, DOI: 10.13328/j.cnki.jos.007240
, CSTR: 32375.14.jos.007240
Abstract:
2025,36(1):371-398
, DOI: 10.13328/j.cnki.jos.007231
, CSTR: 32375.14.jos.007231
Abstract:
2025,36(1):424-445
, DOI: 10.13328/j.cnki.jos.007225
, CSTR: 32375.14.jos.007225
Abstract:
2025,36(1):152-183
, DOI: 10.13328/j.cnki.jos.007217
, CSTR: 32375.14.jos.007217
Abstract:
2025,36(1):107-151
, DOI: 10.13328/j.cnki.jos.007212
, CSTR: 32375.14.jos.007212
Abstract:
2025,36(1):341-370
, DOI: 10.13328/j.cnki.jos.007196
, CSTR: 32375.14.jos.007196
Abstract:
2025,36(1):321-340
, DOI: 10.13328/j.cnki.jos.007194
, CSTR: 32375.14.jos.007194
Abstract:
2025,36(1):289-320
, DOI: 10.13328/j.cnki.jos.007193
, CSTR: 32375.14.jos.007193
Abstract:
2025,36(1):47-78
, DOI: 10.13328/j.cnki.jos.007190
, CSTR: 32375.14.jos.007190
Abstract:
2025,36(1):253-288
, DOI: 10.13328/j.cnki.jos.007182
, CSTR: 32375.14.jos.007182
Abstract:
2025,36(1):203-252
, DOI: 10.13328/j.cnki.jos.007162
, CSTR: 32375.14.jos.007162
Abstract:
2025,36(1):27-46
, DOI: 10.13328/j.cnki.jos.007157
, CSTR: 32375.14.jos.007157
Abstract:
2025,36(1):184-202
, DOI: 10.13328/j.cnki.jos.007148
, CSTR: 32375.14.jos.007148
Abstract:
2025,36(1):1-26
, DOI: 10.13328/j.cnki.jos.007143
, CSTR: 32375.14.jos.007143
Abstract:
2025,36(1):79-106
, DOI: 10.13328/j.cnki.jos.007141
, CSTR: 32375.14.jos.007141
Abstract:
2025,36(1):399-423
, DOI: 10.13328/j.cnki.jos.007048
, CSTR: 32375.14.jos.007048
Abstract:
2024,35(12):5487-5508
, DOI: 10.13328/j.cnki.jos.007232
, CSTR: 32375.14.jos.007232
Abstract:
2024,35(12):5582-5608
, DOI: 10.13328/j.cnki.jos.007089
, CSTR: 32375.14.jos.007089
Abstract:
2024,35(9):4069-4099
, DOI: 10.13328/j.cnki.jos.007138
, CSTR: 32375.14.jos.007138
Abstract:
2024,35(9):4038-4068
, DOI: 10.13328/j.cnki.jos.007127
, CSTR: 32375.14.jos.007127
Abstract:
2024,35(9):4013-4037
, DOI: 10.13328/j.cnki.jos.007126
, CSTR: 32375.14.jos.007126
Abstract:
2024,35(8):3980-4010
, DOI: 10.13328/j.cnki.jos.007095
, CSTR: 32375.14.jos.007095
Abstract:
2024,35(8):3923-3951
, DOI: 10.13328/j.cnki.jos.007093
, CSTR: 32375.14.jos.007093
Abstract:
2024,35(8):3897-3922
, DOI: 10.13328/j.cnki.jos.007087
, CSTR: 32375.14.jos.007087
Abstract:
2024,35(8):3785-3808
, DOI: 10.13328/j.cnki.jos.007080
, CSTR: 32375.14.jos.007080
Abstract:
2024,35(8):3752-3784
, DOI: 10.13328/j.cnki.jos.007059
, CSTR: 32375.14.jos.007059
Abstract:
2024,35(8):3952-3979
, DOI: 10.13328/j.cnki.jos.007045
, CSTR: 32375.14.jos.007045
Abstract:
2024,35(8):3843-3877
, DOI: 10.13328/j.cnki.jos.007042
, CSTR: 32375.14.jos.007042
Abstract:
2024,35(7):3265-3290
, DOI: 10.13328/j.cnki.jos.007112
, CSTR: 32375.14.jos.007112
Abstract:
2024,35(7):3497-3530
, DOI: 10.13328/j.cnki.jos.007053
, CSTR: 32375.14.jos.007053
Abstract:
2024,35(6):2687-2723
, DOI: 10.13328/j.cnki.jos.007099
, CSTR: 32375.14.jos.007099
Abstract:
2024,35(5):2235-2267
, DOI: 10.13328/j.cnki.jos.007046
, CSTR: 32375.14.jos.007046
Abstract:
2024,35(5):2268-2288
, DOI: 10.13328/j.cnki.jos.007038
, CSTR: 32375.14.jos.007038
Abstract:
2024,35(5):2379-2399
, DOI: 10.13328/j.cnki.jos.006804
, CSTR: 32375.14.jos.006804
Abstract:
2024,35(4):1682-1702
, DOI: 10.13328/j.cnki.jos.007012
, CSTR: 32375.14.jos.007012
Abstract:
2024,35(4):1618-1650
, DOI: 10.13328/j.cnki.jos.007011
, CSTR: 32375.14.jos.007011
Abstract:
2024,35(3):1207-1230
, DOI: 10.13328/j.cnki.jos.007075
, CSTR: 32375.14.jos.007075
Abstract:
2024,35(3):1231-1256
, DOI: 10.13328/j.cnki.jos.007061
, CSTR: 32375.14.jos.007061
Abstract:
2024,35(3):1377-1402
, DOI: 10.13328/j.cnki.jos.006807
, CSTR: 32375.14.jos.006807
Abstract:
2024,35(2):711-738
, DOI: 10.13328/j.cnki.jos.007006
, CSTR: 32375.14.jos.007006
Abstract:
2024,35(2):927-974
, DOI: 10.13328/j.cnki.jos.006996
, CSTR: 32375.14.jos.006996
Abstract:
2024,35(2):773-799
, DOI: 10.13328/j.cnki.jos.006985
, CSTR: 32375.14.jos.006985
Abstract:
2024,35(2):1028-1047
, DOI: 10.13328/j.cnki.jos.006984
, CSTR: 32375.14.jos.006984
Abstract:
2024,35(2):629-674
, DOI: 10.13328/j.cnki.jos.006983
, CSTR: 32375.14.jos.006983
Abstract:
2024,35(2):604-628
, DOI: 10.13328/j.cnki.jos.006981
, CSTR: 32375.14.jos.006981
Abstract:
2024,35(2):581-603
, DOI: 10.13328/j.cnki.jos.006975
, CSTR: 32375.14.jos.006975
Abstract:
2024,35(2):828-851
, DOI: 10.13328/j.cnki.jos.006954
, CSTR: 32375.14.jos.006954
Abstract:
2024,35(2):532-580
, DOI: 10.13328/j.cnki.jos.006953
, CSTR: 32375.14.jos.006953
Abstract:
2024,35(2):899-926
, DOI: 10.13328/j.cnki.jos.006952
, CSTR: 32375.14.jos.006952
Abstract:
2024,35(2):800-827
, DOI: 10.13328/j.cnki.jos.006950
, CSTR: 32375.14.jos.006950
Abstract:
2024,35(2):975-1009
, DOI: 10.13328/j.cnki.jos.006946
, CSTR: 32375.14.jos.006946
Abstract:
2024,35(2):513-531
, DOI: 10.13328/j.cnki.jos.006944
, CSTR: 32375.14.jos.006944
Abstract:
2024,35(2):1010-1027
, DOI: 10.13328/j.cnki.jos.006943
, CSTR: 32375.14.jos.006943
Abstract:
2024,35(2):872-898
, DOI: 10.13328/j.cnki.jos.006942
, CSTR: 32375.14.jos.006942
Abstract:
2024,35(2):852-871
, DOI: 10.13328/j.cnki.jos.006938
, CSTR: 32375.14.jos.006938
Abstract:
2024,35(2):675-710
, DOI: 10.13328/j.cnki.jos.006933
, CSTR: 32375.14.jos.006933
Abstract:
2024,35(1):19-37
, DOI: 10.13328/j.cnki.jos.006935
, CSTR: 32375.14.jos.006935
Abstract:
2024,35(1):430-454
, DOI: 10.13328/j.cnki.jos.006917
, CSTR: 32375.14.jos.006917
Abstract:
2024,35(1):405-429
, DOI: 10.13328/j.cnki.jos.006916
, CSTR: 32375.14.jos.006916
Abstract:
2024,35(1):236-265
, DOI: 10.13328/j.cnki.jos.006913
, CSTR: 32375.14.jos.006913
Abstract:
2024,35(1):356-404
, DOI: 10.13328/j.cnki.jos.006911
, CSTR: 32375.14.jos.006911
Abstract:
2024,35(1):1-18
, DOI: 10.13328/j.cnki.jos.006908
, CSTR: 32375.14.jos.006908
Abstract:
2024,35(1):87-117
, DOI: 10.13328/j.cnki.jos.006902
, CSTR: 32375.14.jos.006902
Abstract:
2024,35(1):309-332
, DOI: 10.13328/j.cnki.jos.006898
, CSTR: 32375.14.jos.006898
Abstract:
2024,35(1):333-355
, DOI: 10.13328/j.cnki.jos.006891
, CSTR: 32375.14.jos.006891
Abstract:
2024,35(1):220-235
, DOI: 10.13328/j.cnki.jos.006844
, CSTR: 32375.14.jos.006844
Abstract:
2024,35(1):63-86
, DOI: 10.13328/j.cnki.jos.006835
, CSTR: 32375.14.jos.006835
Abstract:
2024,35(1):185-219
, DOI: 10.13328/j.cnki.jos.006834
, CSTR: 32375.14.jos.006834
Abstract:
2024,35(1):136-158
, DOI: 10.13328/j.cnki.jos.006828
, CSTR: 32375.14.jos.006828
Abstract:
2024,35(1):118-135
, DOI: 10.13328/j.cnki.jos.006827
, CSTR: 32375.14.jos.006827
Abstract:
2024,35(1):455-480
, DOI: 10.13328/j.cnki.jos.006822
, CSTR: 32375.14.jos.006822
Abstract:
2024,35(1):288-308
, DOI: 10.13328/j.cnki.jos.006818
, CSTR: 32375.14.jos.006818
Abstract:
2024,35(1):38-62
, DOI: 10.13328/j.cnki.jos.006810
, CSTR: 32375.14.jos.006810
Abstract:
2024,35(1):481-512
, DOI: 10.13328/j.cnki.jos.006777
, CSTR: 32375.14.jos.006777
Abstract:
2024,35(1):159-184
, DOI: 10.13328/j.cnki.jos.006758
, CSTR: 32375.14.jos.006758
Abstract:
2024,35(1):266-287
, DOI: 10.13328/j.cnki.jos.006753
, CSTR: 32375.14.jos.006753
Abstract:
2023,34(12):5597-5613
, DOI: 10.13328/j.cnki.jos.006806
Abstract:
2023,34(12):5862-5886
, DOI: 10.13328/j.cnki.jos.006800
Abstract:
2023,34(12):5527-5551
, DOI: 10.13328/j.cnki.jos.006765
Abstract:
2023,34(12):5822-5847
, DOI: 10.13328/j.cnki.jos.006764
Abstract:
2023,34(12):5501-5526
, DOI: 10.13328/j.cnki.jos.006749
Abstract:
2023,34(11):5058-5083
, DOI: 10.13328/j.cnki.jos.006735
Abstract:
2023,34(11):4989-5007
, DOI: 10.13328/j.cnki.jos.006684
Abstract:
2023,34(11):5008-5028
, DOI: 10.13328/j.cnki.jos.006679
Abstract:
2023,34(11):5330-5354
, DOI: 10.13328/j.cnki.jos.006669
Abstract:
2023,34(10):4439-4462
, DOI: 10.13328/j.cnki.jos.006884
Abstract:
2023,34(10):4870-4915
, DOI: 10.13328/j.cnki.jos.006680
Abstract:
2023,34(10):4607-4635
, DOI: 10.13328/j.cnki.jos.006677
Abstract:
2023,34(10):4743-4771
, DOI: 10.13328/j.cnki.jos.006672
Abstract:
2023,34(10):4772-4803
, DOI: 10.13328/j.cnki.jos.006671
Abstract:
2023,34(10):4636-4660
, DOI: 10.13328/j.cnki.jos.006666
Abstract:
2023,34(9):4310-4335
, DOI: 10.13328/j.cnki.jos.006657
Abstract:
2023,34(9):4167-4194
, DOI: 10.13328/j.cnki.jos.006650
Abstract:
2023,34(8):3905-3923
, DOI: 10.13328/j.cnki.jos.006645
Abstract:
2023,34(8):3870-3890
, DOI: 10.13328/j.cnki.jos.006635
Abstract:
2023,34(8):3774-3792
, DOI: 10.13328/j.cnki.jos.006592
Abstract:
2023,34(8):3605-3636
, DOI: 10.13328/j.cnki.jos.006558
Abstract:
2023,34(7):3134-3166
, DOI: 10.13328/j.cnki.jos.006863
Abstract:
2023,34(6):2606-2627
, DOI: 10.13328/j.cnki.jos.006850
Abstract:
2023,34(6):2833-2864
, DOI: 10.13328/j.cnki.jos.006658
Abstract:
2023,34(5):2365-2391
, DOI: 10.13328/j.cnki.jos.006808
Abstract:
2023,34(5):2000-2023
, DOI: 10.13328/j.cnki.jos.006774
Abstract:
2023,34(5):2300-2316
, DOI: 10.13328/j.cnki.jos.006485
Abstract:
2023,34(4):1944-1961
, DOI: 10.13328/j.cnki.jos.006725
Abstract:
2023,34(4):1711-1731
, DOI: 10.13328/j.cnki.jos.006700
Abstract:
2023,34(4):1879-1906
, DOI: 10.13328/j.cnki.jos.006692
Abstract:
2023,34(3):1168-1192
, DOI: 10.13328/j.cnki.jos.006786
Abstract:
2023,34(3):1236-1258
, DOI: 10.13328/j.cnki.jos.006785
Abstract:
2023,34(3):1470-1490
, DOI: 10.13328/j.cnki.jos.006778
Abstract:
2023,34(3):1396-1424
, DOI: 10.13328/j.cnki.jos.006751
Abstract:
2023,34(3):1330-1364
, DOI: 10.13328/j.cnki.jos.006717
Abstract:
2023,34(3):1365-1395
, DOI: 10.13328/j.cnki.jos.006715
Abstract:
2023,34(2):582-624
, DOI: 10.13328/j.cnki.jos.006714
Abstract:
2023,34(2):761-785
, DOI: 10.13328/j.cnki.jos.006713
Abstract:
2023,34(2):565-581
, DOI: 10.13328/j.cnki.jos.006711
Abstract:
2023,34(2):690-711
, DOI: 10.13328/j.cnki.jos.006709
Abstract:
2023,34(2):985-1006
, DOI: 10.13328/j.cnki.jos.006707
Abstract:
2023,34(2):733-760
, DOI: 10.13328/j.cnki.jos.006706
Abstract:
2023,34(2):539-564
, DOI: 10.13328/j.cnki.jos.006705
Abstract:
2023,34(2):509-538
, DOI: 10.13328/j.cnki.jos.006704
Abstract:
2023,34(2):712-732
, DOI: 10.13328/j.cnki.jos.006699
Abstract:
2023,34(2):625-654
, DOI: 10.13328/j.cnki.jos.006696
Abstract:
2023,34(2):964-984
, DOI: 10.13328/j.cnki.jos.006693
Abstract:
2023,34(1):404-420
, DOI: 10.13328/j.cnki.jos.006513
Abstract:
2023,34(1):351-380
, DOI: 10.13328/j.cnki.jos.006510
Abstract:
2023,34(1):50-102
, DOI: 10.13328/j.cnki.jos.006503
Abstract:
2023,34(1):463-488
, DOI: 10.13328/j.cnki.jos.006502
Abstract:
2023,34(1):381-403
, DOI: 10.13328/j.cnki.jos.006501
Abstract:
2023,34(1):171-196
, DOI: 10.13328/j.cnki.jos.006500
Abstract:
2023,34(1):444-462
, DOI: 10.13328/j.cnki.jos.006488
Abstract:
2023,34(1):489-508
, DOI: 10.13328/j.cnki.jos.006436
Abstract:
2023,34(1):334-350
, DOI: 10.13328/j.cnki.jos.006435
Abstract:
2023,34(1):421-443
, DOI: 10.13328/j.cnki.jos.006434
Abstract:
2023,34(1):150-170
, DOI: 10.13328/j.cnki.jos.006431
Abstract:
2023,34(1):277-311
, DOI: 10.13328/j.cnki.jos.006429
Abstract:
2023,34(1):130-149
, DOI: 10.13328/j.cnki.jos.006425
Abstract:
2023,34(1):33-49
, DOI: 10.13328/j.cnki.jos.006421
Abstract:
2023,34(1):255-276
, DOI: 10.13328/j.cnki.jos.006420
Abstract:
2023,34(1):230-254
, DOI: 10.13328/j.cnki.jos.006415
Abstract:
2023,34(1):197-229
, DOI: 10.13328/j.cnki.jos.006411
Abstract:
2023,34(1):103-129
, DOI: 10.13328/j.cnki.jos.006409
Abstract:
2023,34(1):1-32
, DOI: 10.13328/j.cnki.jos.006402
Abstract:
2023,34(1):312-333
, DOI: 10.13328/j.cnki.jos.006395
Abstract:
2022,33(12):4668-4687
, DOI: 10.13328/j.cnki.jos.006544
Abstract:
2022,33(12):4644-4667
, DOI: 10.13328/j.cnki.jos.006522
Abstract:
2022,33(12):4616-4643
, DOI: 10.13328/j.cnki.jos.006521
Abstract:
2022,33(12):4590-4615
, DOI: 10.13328/j.cnki.jos.006518
Abstract:
2022,33(11):4107-4136
, DOI: 10.13328/j.cnki.jos.006539
Abstract:
2022,33(11):4084-4106
, DOI: 10.13328/j.cnki.jos.006528
Abstract:
2022,33(11):4334-4355
, DOI: 10.13328/j.cnki.jos.006524
Abstract:
2022,33(11):4217-4238
, DOI: 10.13328/j.cnki.jos.006391
Abstract:
2022,33(11):4173-4191
, DOI: 10.13328/j.cnki.jos.006390
Abstract:
2022,33(11):4192-4216
, DOI: 10.13328/j.cnki.jos.006385
Abstract:
2022,33(10):3635-3655
, DOI: 10.13328/j.cnki.jos.006631
Abstract:
2022,33(10):3582-3618
, DOI: 10.13328/j.cnki.jos.006626
Abstract:
2022,33(10):3673-3699
, DOI: 10.13328/j.cnki.jos.006372
Abstract:
2022,33(10):3732-3753
, DOI: 10.13328/j.cnki.jos.006365
Abstract:
2022,33(9):3370-3390
, DOI: 10.13328/j.cnki.jos.006426
Abstract:
2022,33(8):3059-3085
, DOI: 10.13328/j.cnki.jos.006375
Abstract:
2022,33(8):3015-3034
, DOI: 10.13328/j.cnki.jos.006355
Abstract:
2022,33(7):2411-2446
, DOI: 10.13328/j.cnki.jos.006590
Abstract:
2022,33(7):2367-2410
, DOI: 10.13328/j.cnki.jos.006587
Abstract:
2022,33(7):2729-2752
, DOI: 10.13328/j.cnki.jos.006299
Abstract:
2022,33(6):2288-2311
, DOI: 10.13328/j.cnki.jos.006367
Abstract:
2022,33(5):1612-1634
, DOI: 10.13328/j.cnki.jos.006551
Abstract:
2022,33(5):1800-1816
, DOI: 10.13328/j.cnki.jos.006122
Abstract:
2022,33(4):1244-1266
, DOI: 10.13328/j.cnki.jos.006481
Abstract:
2022,33(4):1287-1314
, DOI: 10.13328/j.cnki.jos.006476
Abstract:
2022,33(3):1057-1092
, DOI: 10.13328/j.cnki.jos.006446
Abstract:
2022,33(2):751-769
, DOI: 10.13328/j.cnki.jos.006370
Abstract:
2022,33(2):598-621
, DOI: 10.13328/j.cnki.jos.006357
Abstract:
2022,33(2):498-523
, DOI: 10.13328/j.cnki.jos.006353
Abstract:
2022,33(1):78-101
, DOI: 10.13328/j.cnki.jos.006345
Abstract:
2022,33(1):297-323
, DOI: 10.13328/j.cnki.jos.006344
Abstract:
2022,33(1):193-210
, DOI: 10.13328/j.cnki.jos.006342
Abstract:
2022,33(1):274-296
, DOI: 10.13328/j.cnki.jos.006338
Abstract:
2022,33(1):55-77
, DOI: 10.13328/j.cnki.jos.006337
Abstract:
2022,33(1):356-376
, DOI: 10.13328/j.cnki.jos.006334
Abstract:
2022,33(1):171-192
, DOI: 10.13328/j.cnki.jos.006323
Abstract:
2022,33(1):150-170
, DOI: 10.13328/j.cnki.jos.006311
Abstract:
2022,33(1):129-149
, DOI: 10.13328/j.cnki.jos.006306
Abstract:
2022,33(1):102-128
, DOI: 10.13328/j.cnki.jos.006304
Abstract:
2022,33(1):254-273
, DOI: 10.13328/j.cnki.jos.006303
Abstract:
2022,33(1):26-54
, DOI: 10.13328/j.cnki.jos.006292
Abstract:
2022,33(1):324-355
, DOI: 10.13328/j.cnki.jos.006280
Abstract:
2022,33(1):1-25
, DOI: 10.13328/j.cnki.jos.006279
Abstract:
2022,33(1):233-253
, DOI: 10.13328/j.cnki.jos.006219
Abstract:
2022,33(1):211-232
, DOI: 10.13328/j.cnki.jos.006218
Abstract:
2021,32(12):3992-4024
, DOI: 10.13328/j.cnki.jos.006490
Abstract:
2021,32(10):3051-3067
, DOI: 10.13328/j.cnki.jos.006217
Abstract:
2021,32(10):3176-3202
, DOI: 10.13328/j.cnki.jos.006203
Abstract:
2021,32(9):2642-2664
, DOI: 10.13328/j.cnki.jos.006275
Abstract:
2021,32(9):2665-2690
, DOI: 10.13328/j.cnki.jos.006274
Abstract:
2021,32(8):2522-2544
, DOI: 10.13328/j.cnki.jos.006215
Abstract:
2021,32(8):2469-2504
, DOI: 10.13328/j.cnki.jos.006153
Abstract:
2021,32(7):2260-2286
, DOI: 10.13328/j.cnki.jos.006309
Abstract:
2021,32(7):2078-2102
, DOI: 10.13328/j.cnki.jos.006269
Abstract:
2021,32(7):1926-1956
, DOI: 10.13328/j.cnki.jos.006267
Abstract:
2026,37(5):1885-1886
, DOI: 10.13328/j.cnki.jos.007546
, CSTR: 32375.14.jos.007546
Abstract:
2026,37(5):1919-1935
, DOI: 10.13328/j.cnki.jos.007545
, CSTR: 32375.14.jos.007545
Abstract:
2026,37(5):1950-1963
, DOI: 10.13328/j.cnki.jos.007544
, CSTR: 32375.14.jos.007544
Abstract:
2026,37(5):1936-1949
, DOI: 10.13328/j.cnki.jos.007543
, CSTR: 32375.14.jos.007543
Abstract:
2026,37(5):1887-1902
, DOI: 10.13328/j.cnki.jos.007542
, CSTR: 32375.14.jos.007542
Abstract:
2026,37(5):1903-1918
, DOI: 10.13328/j.cnki.jos.007541
, CSTR: 32375.14.jos.007541
Abstract:
2026,37(5):2006-2023
, DOI: 10.13328/j.cnki.jos.007540
, CSTR: 32375.14.jos.007540
Abstract:
2026,37(5):1964-1981
, DOI: 10.13328/j.cnki.jos.007537
, CSTR: 32375.14.jos.007537
Abstract:
2026,37(4):1447-1448
, DOI: 10.13328/j.cnki.jos.007531
, CSTR: 32375.14.jos.007531
Abstract:
2026,37(4):1575-1590
, DOI: 10.13328/j.cnki.jos.007528
, CSTR: 32375.14.jos.007528
Abstract:
2026,37(4):1560-1574
, DOI: 10.13328/j.cnki.jos.007527
, CSTR: 32375.14.jos.007527
Abstract:
2026,37(4):1548-1559
, DOI: 10.13328/j.cnki.jos.007526
, CSTR: 32375.14.jos.007526
Abstract:
2026,37(4):1531-1547
, DOI: 10.13328/j.cnki.jos.007525
, CSTR: 32375.14.jos.007525
Abstract:
2026,37(4):1511-1530
, DOI: 10.13328/j.cnki.jos.007524
, CSTR: 32375.14.jos.007524
Abstract:
2026,37(4):1492-1510
, DOI: 10.13328/j.cnki.jos.007523
, CSTR: 32375.14.jos.007523
Abstract:
2026,37(4):1472-1491
, DOI: 10.13328/j.cnki.jos.007522
, CSTR: 32375.14.jos.007522
Abstract:
2026,37(4):1449-1471
, DOI: 10.13328/j.cnki.jos.007521
, CSTR: 32375.14.jos.007521
Abstract:
2026,37(3):969-970
, DOI: 10.13328/j.cnki.jos.007520
, CSTR: 32375.14.jos.007520
Abstract:
2026,37(3):1084-1103
, DOI: 10.13328/j.cnki.jos.007519
, CSTR: 32375.14.jos.007519
Abstract:
2026,37(3):1021-1036
, DOI: 10.13328/j.cnki.jos.007518
, CSTR: 32375.14.jos.007518
Abstract:
2026,37(3):1006-1020
, DOI: 10.13328/j.cnki.jos.007517
, CSTR: 32375.14.jos.007517
Abstract:
2026,37(3):1104-1120
, DOI: 10.13328/j.cnki.jos.007515
, CSTR: 32375.14.jos.007515
Abstract:
2026,37(3):1121-1142
, DOI: 10.13328/j.cnki.jos.007514
, CSTR: 32375.14.jos.007514
Abstract:
2026,37(3):1058-1083
, DOI: 10.13328/j.cnki.jos.007513
, CSTR: 32375.14.jos.007513
Abstract:
2026,37(3):1037-1057
, DOI: 10.13328/j.cnki.jos.007512
, CSTR: 32375.14.jos.007512
Abstract:
2025,36(10):4403-4404
, DOI: 10.13328/j.cnki.jos.007395
, CSTR: 32375.14.jos.007395
Abstract:
2025,36(10):4483-4506
, DOI: 10.13328/j.cnki.jos.007394
, CSTR: 32375.14.jos.007394
Abstract:
2025,36(10):4430-4443
, DOI: 10.13328/j.cnki.jos.007393
, CSTR: 32375.14.jos.007393
Abstract:
2025,36(10):4417-4429
, DOI: 10.13328/j.cnki.jos.007392
, CSTR: 32375.14.jos.007392
Abstract:
2025,36(10):4507-4524
, DOI: 10.13328/j.cnki.jos.007391
, CSTR: 32375.14.jos.007391
Abstract:
2025,36(10):4444-4460
, DOI: 10.13328/j.cnki.jos.007390
, CSTR: 32375.14.jos.007390
Abstract:
2025,36(10):4461-4482
, DOI: 10.13328/j.cnki.jos.007389
, CSTR: 32375.14.jos.007389
Abstract:
2025,36(10):4405-4416
, DOI: 10.13328/j.cnki.jos.007388
, CSTR: 32375.14.jos.007388
Abstract:
2025,36(9):3917-3918
, DOI: 10.13328/j.cnki.jos.007361
, CSTR: 32375.14.jos.007361
Abstract:
2025,36(9):3985-4005
, DOI: 10.13328/j.cnki.jos.007360
, CSTR: 32375.14.jos.007360
Abstract:
2025,36(9):3970-3984
, DOI: 10.13328/j.cnki.jos.007359
, CSTR: 32375.14.jos.007359
Abstract:
2025,36(9):3954-3969
, DOI: 10.13328/j.cnki.jos.007358
, CSTR: 32375.14.jos.007358
Abstract:
2025,36(9):3937-3953
, DOI: 10.13328/j.cnki.jos.007357
, CSTR: 32375.14.jos.007357
Abstract:
2025,36(9):3919-3936
, DOI: 10.13328/j.cnki.jos.007292
, CSTR: 32375.14.jos.007292
Abstract:
2025,36(8):3429-3430
, DOI: 10.13328/j.cnki.jos.007355
, CSTR: 32375.14.jos.007355
Abstract:
2025,36(8):3637-3654
, DOI: 10.13328/j.cnki.jos.007354
, CSTR: 32375.14.jos.007354
Abstract:
2025,36(8):3587-3603
, DOI: 10.13328/j.cnki.jos.007352
, CSTR: 32375.14.jos.007352
Abstract:
2025,36(8):3570-3586
, DOI: 10.13328/j.cnki.jos.007351
, CSTR: 32375.14.jos.007351
Abstract:
2025,36(8):3554-3569
, DOI: 10.13328/j.cnki.jos.007350
, CSTR: 32375.14.jos.007350
Abstract:
2025,36(8):3531-3553
, DOI: 10.13328/j.cnki.jos.007349
, CSTR: 32375.14.jos.007349
Abstract:
2025,36(8):3512-3530
, DOI: 10.13328/j.cnki.jos.007348
, CSTR: 32375.14.jos.007348
Abstract:
2025,36(8):3494-3511
, DOI: 10.13328/j.cnki.jos.007347
, CSTR: 32375.14.jos.007347
Abstract:
2025,36(8):3477-3493
, DOI: 10.13328/j.cnki.jos.007346
, CSTR: 32375.14.jos.007346
Abstract:
2025,36(8):3462-3476
, DOI: 10.13328/j.cnki.jos.007345
, CSTR: 32375.14.jos.007345
Abstract:
2025,36(8):3444-3461
, DOI: 10.13328/j.cnki.jos.007344
, CSTR: 32375.14.jos.007344
Abstract:
2025,36(8):3431-3443
, DOI: 10.13328/j.cnki.jos.007343
, CSTR: 32375.14.jos.007343
Abstract:
2025,36(7):2927-2928
, DOI: 10.13328/j.cnki.jos.007342
, CSTR: 32375.14.jos.007342
Abstract:
2025,36(7):3134-3150
, DOI: 10.13328/j.cnki.jos.007341
, CSTR: 32375.14.jos.007341
Abstract:
2025,36(7):3109-3133
, DOI: 10.13328/j.cnki.jos.007340
, CSTR: 32375.14.jos.007340
Abstract:
2025,36(7):3087-3108
, DOI: 10.13328/j.cnki.jos.007339
, CSTR: 32375.14.jos.007339
Abstract:
2025,36(7):3041-3086
, DOI: 10.13328/j.cnki.jos.007338
, CSTR: 32375.14.jos.007338
Abstract:
2025,36(7):3022-3040
, DOI: 10.13328/j.cnki.jos.007336
, CSTR: 32375.14.jos.007336
Abstract:
2025,36(7):3003-3021
, DOI: 10.13328/j.cnki.jos.007335
, CSTR: 32375.14.jos.007335
Abstract:
2025,36(7):2947-2963
, DOI: 10.13328/j.cnki.jos.007333
, CSTR: 32375.14.jos.007333
Abstract:
2025,36(7):2929-2946
, DOI: 10.13328/j.cnki.jos.007332
, CSTR: 32375.14.jos.007332
Abstract:
2025,36(6):2401-2403
, DOI: 10.13328/j.cnki.jos.007331
, CSTR: 32375.14.jos.007331
Abstract:
2025,36(6):2515-2534
, DOI: 10.13328/j.cnki.jos.007330
, CSTR: 32375.14.jos.007330
Abstract:
2025,36(6):2501-2514
, DOI: 10.13328/j.cnki.jos.007329
, CSTR: 32375.14.jos.007329
Abstract:
2025,36(6):2432-2452
, DOI: 10.13328/j.cnki.jos.007327
, CSTR: 32375.14.jos.007327
Abstract:
2025,36(6):2558-2575
, DOI: 10.13328/j.cnki.jos.007326
, CSTR: 32375.14.jos.007326
Abstract:
2025,36(6):2477-2500
, DOI: 10.13328/j.cnki.jos.007325
, CSTR: 32375.14.jos.007325
Abstract:
2025,36(6):2535-2557
, DOI: 10.13328/j.cnki.jos.007324
, CSTR: 32375.14.jos.007324
Abstract:
2025,36(3):939-940
, DOI: 10.13328/j.cnki.jos.007284
, CSTR: 32375.14.jos.007284
Abstract:
2025,36(3):1084-1106
, DOI: 10.13328/j.cnki.jos.007283
, CSTR: 32375.14.jos.007283
Abstract:
2025,36(3):1022-1039
, DOI: 10.13328/j.cnki.jos.007282
, CSTR: 32375.14.jos.007282
Abstract:
2025,36(3):995-1021
, DOI: 10.13328/j.cnki.jos.007281
, CSTR: 32375.14.jos.007281
Abstract:
2025,36(3):981-994
, DOI: 10.13328/j.cnki.jos.007280
, CSTR: 32375.14.jos.007280
Abstract:
2025,36(3):1065-1083
, DOI: 10.13328/j.cnki.jos.007279
, CSTR: 32375.14.jos.007279
Abstract:
2025,36(3):962-980
, DOI: 10.13328/j.cnki.jos.007278
, CSTR: 32375.14.jos.007278
Abstract:
2025,36(3):941-961
, DOI: 10.13328/j.cnki.jos.007277
, CSTR: 32375.14.jos.007277
Abstract:
2024,35(9):4011-4012
, DOI: 10.13328/j.cnki.jos.007140
, CSTR: 32375.14.jos.007140
Abstract:
2024,35(9):4310-4323
, DOI: 10.13328/j.cnki.jos.007139
, CSTR: 32375.14.jos.007139
Abstract:
2024,35(9):4287-4309
, DOI: 10.13328/j.cnki.jos.007137
, CSTR: 32375.14.jos.007137
Abstract:
2024,35(9):4265-4286
, DOI: 10.13328/j.cnki.jos.007136
, CSTR: 32375.14.jos.007136
Abstract:
2024,35(9):4242-4264
, DOI: 10.13328/j.cnki.jos.007135
, CSTR: 32375.14.jos.007135
Abstract:
2024,35(9):4218-4241
, DOI: 10.13328/j.cnki.jos.007134
, CSTR: 32375.14.jos.007134
Abstract:
2024,35(9):4193-4217
, DOI: 10.13328/j.cnki.jos.007133
, CSTR: 32375.14.jos.007133
Abstract:
2024,35(9):4179-4192
, DOI: 10.13328/j.cnki.jos.007132
, CSTR: 32375.14.jos.007132
Abstract:
2024,35(9):4160-4178
, DOI: 10.13328/j.cnki.jos.007131
, CSTR: 32375.14.jos.007131
Abstract:
2024,35(9):4141-4159
, DOI: 10.13328/j.cnki.jos.007130
, CSTR: 32375.14.jos.007130
Abstract:
2024,35(9):4123-4140
, DOI: 10.13328/j.cnki.jos.007129
, CSTR: 32375.14.jos.007129
Abstract:
2024,35(9):4100-4122
, DOI: 10.13328/j.cnki.jos.007128
, CSTR: 32375.14.jos.007128
Abstract:
2024,35(8):3551-3552
, DOI: 10.13328/j.cnki.jos.007125
, CSTR: 32375.14.jos.007125
Abstract:
2024,35(8):3698-3720
, DOI: 10.13328/j.cnki.jos.007124
, CSTR: 32375.14.jos.007124
Abstract:
2024,35(8):3684-3697
, DOI: 10.13328/j.cnki.jos.007122
, CSTR: 32375.14.jos.007122
Abstract:
2024,35(8):3668-3683
, DOI: 10.13328/j.cnki.jos.007121
, CSTR: 32375.14.jos.007121
Abstract:
2024,35(8):3647-3667
, DOI: 10.13328/j.cnki.jos.007120
, CSTR: 32375.14.jos.007120
Abstract:
2024,35(8):3626-3646
, DOI: 10.13328/j.cnki.jos.007119
, CSTR: 32375.14.jos.007119
Abstract:
2024,35(8):3610-3625
, DOI: 10.13328/j.cnki.jos.007118
, CSTR: 32375.14.jos.007118
Abstract:
2024,35(8):3591-3609
, DOI: 10.13328/j.cnki.jos.007117
, CSTR: 32375.14.jos.007117
Abstract:
2024,35(8):3577-3590
, DOI: 10.13328/j.cnki.jos.007116
, CSTR: 32375.14.jos.007116
Abstract:
2024,35(8):3553-3576
, DOI: 10.13328/j.cnki.jos.007115
, CSTR: 32375.14.jos.007115
Abstract:
2024,35(7):3069-3070
, DOI: 10.13328/j.cnki.jos.007114
, CSTR: 32375.14.jos.007114
Abstract:
2024,35(7):3093-3114
, DOI: 10.13328/j.cnki.jos.007113
, CSTR: 32375.14.jos.007113
Abstract:
2024,35(7):3245-3264
, DOI: 10.13328/j.cnki.jos.007111
, CSTR: 32375.14.jos.007111
Abstract:
2024,35(7):3071-3092
, DOI: 10.13328/j.cnki.jos.007110
, CSTR: 32375.14.jos.007110
Abstract:
2024,35(7):3227-3244
, DOI: 10.13328/j.cnki.jos.007109
, CSTR: 32375.14.jos.007109
Abstract:
2024,35(7):3204-3226
, DOI: 10.13328/j.cnki.jos.007108
, CSTR: 32375.14.jos.007108
Abstract:
2024,35(7):3180-3203
, DOI: 10.13328/j.cnki.jos.007107
, CSTR: 32375.14.jos.007107
Abstract:
2024,35(7):3162-3179
, DOI: 10.13328/j.cnki.jos.007106
, CSTR: 32375.14.jos.007106
Abstract:
2024,35(7):3141-3161
, DOI: 10.13328/j.cnki.jos.007105
, CSTR: 32375.14.jos.007105
Abstract:
2024,35(7):3115-3140
, DOI: 10.13328/j.cnki.jos.007104
, CSTR: 32375.14.jos.007104
Abstract:
2024,35(6):2583-2584
, DOI: 10.13328/j.cnki.jos.007103
, CSTR: 32375.14.jos.007103
Abstract:
2024,35(6):2668-2686
, DOI: 10.13328/j.cnki.jos.007102
, CSTR: 32375.14.jos.007102
Abstract:
2024,35(6):2724-2752
, DOI: 10.13328/j.cnki.jos.007101
, CSTR: 32375.14.jos.007101
Abstract:
2024,35(6):2608-2630
, DOI: 10.13328/j.cnki.jos.007100
, CSTR: 32375.14.jos.007100
Abstract:
2024,35(6):2648-2667
, DOI: 10.13328/j.cnki.jos.007098
, CSTR: 32375.14.jos.007098
Abstract:
2024,35(6):2631-2647
, DOI: 10.13328/j.cnki.jos.007097
, CSTR: 32375.14.jos.007097
Abstract:
2024,35(6):2585-2607
, DOI: 10.13328/j.cnki.jos.007096
, CSTR: 32375.14.jos.007096
Abstract:
2024,35(5):2099-2100
, DOI: 10.13328/j.cnki.jos.007030
, CSTR: 32375.14.jos.007030
Abstract:
2024,35(5):2176-2191
, DOI: 10.13328/j.cnki.jos.007029
, CSTR: 32375.14.jos.007029
Abstract:
2024,35(5):2192-2207
, DOI: 10.13328/j.cnki.jos.007028
, CSTR: 32375.14.jos.007028
Abstract:
2024,35(5):2165-2175
, DOI: 10.13328/j.cnki.jos.007027
, CSTR: 32375.14.jos.007027
Abstract:
2024,35(5):2208-2219
, DOI: 10.13328/j.cnki.jos.007026
, CSTR: 32375.14.jos.007026
Abstract:
2024,35(5):2133-2149
, DOI: 10.13328/j.cnki.jos.007025
, CSTR: 32375.14.jos.007025
Abstract:
2024,35(5):2150-2164
, DOI: 10.13328/j.cnki.jos.007024
, CSTR: 32375.14.jos.007024
Abstract:
2024,35(5):2220-2234
, DOI: 10.13328/j.cnki.jos.007023
, CSTR: 32375.14.jos.007023
Abstract:
2024,35(5):2101-2119
, DOI: 10.13328/j.cnki.jos.007022
, CSTR: 32375.14.jos.007022
Abstract:
2024,35(5):2120-2132
, DOI: 10.13328/j.cnki.jos.007021
, CSTR: 32375.14.jos.007021
Abstract:
2024,35(4):1585-1586
, DOI: 10.13328/j.cnki.jos.007020
, CSTR: 32375.14.jos.007020
Abstract:
2024,35(4):1774-1789
, DOI: 10.13328/j.cnki.jos.007019
, CSTR: 32375.14.jos.007019
Abstract:
2024,35(4):1751-1773
, DOI: 10.13328/j.cnki.jos.007018
, CSTR: 32375.14.jos.007018
Abstract:
2024,35(4):1732-1750
, DOI: 10.13328/j.cnki.jos.007017
, CSTR: 32375.14.jos.007017
Abstract:
2024,35(4):1601-1617
, DOI: 10.13328/j.cnki.jos.007016
, CSTR: 32375.14.jos.007016
Abstract:
2024,35(4):1703-1716
, DOI: 10.13328/j.cnki.jos.007015
, CSTR: 32375.14.jos.007015
Abstract:
2024,35(4):1587-1600
, DOI: 10.13328/j.cnki.jos.007014
, CSTR: 32375.14.jos.007014
Abstract:
2024,35(4):1717-1731
, DOI: 10.13328/j.cnki.jos.007013
, CSTR: 32375.14.jos.007013
Abstract:
2024,35(4):1651-1666
, DOI: 10.13328/j.cnki.jos.007010
, CSTR: 32375.14.jos.007010
Abstract:
2024,35(4):1667-1681
, DOI: 10.13328/j.cnki.jos.007009
, CSTR: 32375.14.jos.007009
Abstract:
2024,35(3):1049-1050
, DOI: 10.13328/j.cnki.jos.007079
, CSTR: 32375.14.jos.007079
Abstract:
2024,35(3):1140-1153
, DOI: 10.13328/j.cnki.jos.007078
, CSTR: 32375.14.jos.007078
Abstract:
2024,35(3):1173-1193
, DOI: 10.13328/j.cnki.jos.007077
, CSTR: 32375.14.jos.007077
Abstract:
2024,35(3):1074-1089
, DOI: 10.13328/j.cnki.jos.007076
, CSTR: 32375.14.jos.007076
Abstract:
2024,35(3):1194-1206
, DOI: 10.13328/j.cnki.jos.007074
, CSTR: 32375.14.jos.007074
Abstract:
2024,35(3):1125-1139
, DOI: 10.13328/j.cnki.jos.007073
, CSTR: 32375.14.jos.007073
Abstract:
2024,35(3):1090-1106
, DOI: 10.13328/j.cnki.jos.007072
, CSTR: 32375.14.jos.007072
Abstract:
2024,35(3):1051-1073
, DOI: 10.13328/j.cnki.jos.007071
, CSTR: 32375.14.jos.007071
Abstract:
2024,35(3):1154-1172
, DOI: 10.13328/j.cnki.jos.007070
, CSTR: 32375.14.jos.007070
Abstract:
2024,35(3):1107-1124
, DOI: 10.13328/j.cnki.jos.007069
, CSTR: 32375.14.jos.007069
Abstract:
2023,34(10):4437-4438
, DOI: 10.13328/j.cnki.jos.006890
Abstract:
2023,34(10):4565-4583
, DOI: 10.13328/j.cnki.jos.006889
Abstract:
2023,34(10):4533-4547
, DOI: 10.13328/j.cnki.jos.006888
Abstract:
2023,34(10):4501-4517
, DOI: 10.13328/j.cnki.jos.006887
Abstract:
2023,34(10):4548-4564
, DOI: 10.13328/j.cnki.jos.006886
Abstract:
2023,34(10):4518-4532
, DOI: 10.13328/j.cnki.jos.006885
Abstract:
2023,34(10):4477-4500
, DOI: 10.13328/j.cnki.jos.006883
Abstract:
2023,34(10):4584-4600
, DOI: 10.13328/j.cnki.jos.006882
Abstract:
2023,34(10):4463-4476
, DOI: 10.13328/j.cnki.jos.006881
Abstract:
2023,34(9):3939-3940
, DOI: 10.13328/j.cnki.jos.006880
Abstract:
2023,34(9):4018-4036
, DOI: 10.13328/j.cnki.jos.006879
Abstract:
2023,34(9):4003-4017
, DOI: 10.13328/j.cnki.jos.006878
Abstract:
2023,34(9):3981-4002
, DOI: 10.13328/j.cnki.jos.006877
Abstract:
2023,34(9):4069-4095
, DOI: 10.13328/j.cnki.jos.006876
Abstract:
2023,34(9):3941-3965
, DOI: 10.13328/j.cnki.jos.006875
Abstract:
2023,34(9):3966-3980
, DOI: 10.13328/j.cnki.jos.006874
Abstract:
2023,34(9):4056-4068
, DOI: 10.13328/j.cnki.jos.006873
Abstract:
2023,34(9):4037-4055
, DOI: 10.13328/j.cnki.jos.006872
Abstract:
2023,34(8):3465-3466
, DOI: 10.13328/j.cnki.jos.006871
Abstract:
2023,34(8):3574-3586
, DOI: 10.13328/j.cnki.jos.006870
Abstract:
2023,34(8):3527-3548
, DOI: 10.13328/j.cnki.jos.006869
Abstract:
2023,34(8):3549-3573
, DOI: 10.13328/j.cnki.jos.006868
Abstract:
2023,34(8):3467-3484
, DOI: 10.13328/j.cnki.jos.006867
Abstract:
2023,34(8):3507-3526
, DOI: 10.13328/j.cnki.jos.006866
Abstract:
2023,34(8):3485-3506
, DOI: 10.13328/j.cnki.jos.006865
Abstract:
2023,34(7):2979-2980
, DOI: 10.13328/j.cnki.jos.006864
Abstract:
2023,34(7):3043-3063
, DOI: 10.13328/j.cnki.jos.006862
Abstract:
2023,34(7):3080-3098
, DOI: 10.13328/j.cnki.jos.006861
Abstract:
2023,34(7):3116-3133
, DOI: 10.13328/j.cnki.jos.006860
Abstract:
2023,34(7):3099-3115
, DOI: 10.13328/j.cnki.jos.006859
Abstract:
2023,34(7):3022-3042
, DOI: 10.13328/j.cnki.jos.006858
Abstract:
2023,34(7):2981-3001
, DOI: 10.13328/j.cnki.jos.006857
Abstract:
2023,34(7):3064-3079
, DOI: 10.13328/j.cnki.jos.006856
Abstract:
2023,34(7):3002-3021
, DOI: 10.13328/j.cnki.jos.006855
Abstract:
2023,34(6):2507-2508
, DOI: 10.13328/j.cnki.jos.006854
Abstract:
2023,34(6):2562-2585
, DOI: 10.13328/j.cnki.jos.006853
Abstract:
2023,34(6):2628-2640
, DOI: 10.13328/j.cnki.jos.006852
Abstract:
2023,34(6):2586-2605
, DOI: 10.13328/j.cnki.jos.006851
Abstract:
2023,34(6):2543-2561
, DOI: 10.13328/j.cnki.jos.006849
Abstract:
2023,34(6):2526-2542
, DOI: 10.13328/j.cnki.jos.006848
Abstract:
2023,34(6):2641-2668
, DOI: 10.13328/j.cnki.jos.006847
Abstract:
2023,34(6):2509-2525
, DOI: 10.13328/j.cnki.jos.006846
Abstract:
2023,34(5):1997-1999
, DOI: 10.13328/j.cnki.jos.006776
Abstract:
2023,34(5):2083-2100
, DOI: 10.13328/j.cnki.jos.006775
Abstract:
2023,34(5):2152-2169
, DOI: 10.13328/j.cnki.jos.006773
Abstract:
2023,34(5):2134-2151
, DOI: 10.13328/j.cnki.jos.006772
Abstract:
2023,34(5):2068-2082
, DOI: 10.13328/j.cnki.jos.006771
Abstract:
2023,34(5):2024-2034
, DOI: 10.13328/j.cnki.jos.006770
Abstract:
2023,34(5):2116-2133
, DOI: 10.13328/j.cnki.jos.006769
Abstract:
2023,34(5):2051-2067
, DOI: 10.13328/j.cnki.jos.006768
Abstract:
2023,34(5):2035-2050
, DOI: 10.13328/j.cnki.jos.006766
Abstract:
2023,34(3):1007-1009
, DOI: 10.13328/j.cnki.jos.006796
Abstract:
2023,34(3):1109-1125
, DOI: 10.13328/j.cnki.jos.006795
Abstract:
2023,34(3):1087-1108
, DOI: 10.13328/j.cnki.jos.006794
Abstract:
2023,34(3):1065-1086
, DOI: 10.13328/j.cnki.jos.006793
Abstract:
2023,34(3):1277-1291
, DOI: 10.13328/j.cnki.jos.006792
Abstract:
学术社区
External Links
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-4
Address：4# South Fourth Street, Zhong Guan Cun, Beijing 100190,Postal Code：100190
Phone：010-62562563 Fax：010-62562533 Email：jos@iscas.ac.cn
Technical Support：Beijing Qinyun Technology Development Co., Ltd.
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-4
Address：4# South Fourth Street, Zhong Guan Cun, Beijing 100190,Postal Code：100190
Phone：010-62562563 Fax：010-62562533 Email：jos@iscas.ac.cn
Technical Support：Beijing Qinyun Technology Development Co., Ltd.
