| 本文已被：浏览 1090次 下载 1766次 |
码上扫一扫！ |
|
|
| 基于多模态融合的软件缺陷协同分派方法 |
|
谢生龙1,2,3, 李青山1,3, 歹杰1,3, 崔笛1,3
|
|
1.西安电子科技大学 计算机科学与技术学院, 陕西 西安 710126;2.延安大学 数学与计算机科学学院, 陕西 延安 716000;3.西安市智能软件工程重点实验室 (西安电子科技大学), 陕西 西安 710126
|
|
| 摘要: |
| 软件缺陷(bug)分派是将bug报告与适合解决该bug的开发人员进行匹配的过程, 能够使bug得到及时修复. 目前的bug分派研究大多集中于bug报告的文本分类, 但根据帕累托法则, 用以分类的bug报告存在数据分布不均衡现象, 容易对非活跃开发者产生较差的分派效果; 此外, 现有的分类模型忽视了对开发人员的建模且难以挖掘bug与开发人员之间的相关性, 影响了bug分派效能. 为此, 提出一种基于多模态融合的软件缺陷协同分派方法CBT-MF (collaborative bug triaging method based on multimodal fusion). 该方法首先对bug报告进行预处理并构造bug-开发人员二部图; 其次, 为了缓减bug修复记录分布不均衡性的影响, 通过K-means和正负采样的方法对二部图数据进行增强; 为了表征开发者信息, 基于图卷积模型提取二部图节点特征; 最后, 采用内积匹配的方法捕获bug与开发者的相关性, 并通过贝叶斯个性化排序实现bug报告与开发人员的推荐与分派. 在公开数据集上进行全面的实验评估, 实验结果表明, CBT-MF在bug分派方面相较于多个现有先进方法表现出更优越的性能. |
| 关键词: 缺陷分派 不均衡性 多模态融合 图卷积 |
| DOI：10.13328/j.cnki.jos.007267 |
| 分类号:TP311 |
| 基金项目:国家自然科学基金(U21B2015, 62202357); 陕西省自然科学基础研究计划(2023-JC-QN-0744); 陕西省科协青年人才托举计划(20220113); 西安电子科技大学杭州研究院概念验证基金(XJ2023230039) |
|
| Collaborative Bug Triaging Method Based on Multimodal Fusion |
|
XIE Sheng-Long1,2,3, LI Qing-Shan1,3, DAI Jie1,3, CUI Di1,3
|
|
1.School of Computer Science and Technology, Xidian University, Xi’an 710126, China;2.College of Mathematics and Computer Science, Yan’an University, Yan’an 716000, China;3.Xi’an Key Laboratory of Intelligent Software Engineering (Xidian University), Xi’an 710126, China
|
| Abstract: |
| Bug triaging is the process of assigning bug reports to developers suitable for resolving the reported bugs, ensuring timely fixes. Current research in bug triaging mainly focuses on the text classification of bug reports. However, according to the Pareto principle, the data distribution of bug reports used for classification is unbalanced, which may lead to ineffective triaging for inactive developers. Additionally, existing classification models often neglect to model developers and struggle to capture the correlations between bugs and developers, affecting the efficiency of bug triaging. To address these issues, this study proposes a collaborative bug triaging method based on multimodal fusion (CBT-MF). This method first preprocesses bug reports and constructs a bug-developer bipartite graph. To mitigate the impact of the unbalanced distribution of bug fix records, the bipartite graph data is enhanced using K-means clustering and positive-negative sampling. To represent developer information, node features are extracted from the bipartite graph using a graph convolutional network model. Finally, correlations between bugs and developers are captured by matching inner products, and Bayesian personalized ranking (BPR) is utilized for bug report recommendation and triaging. Comprehensive experiments conducted on publicly available datasets demonstrate that CBT-MF outperforms several state-of-the-art methods in bug triaging. |
| Key words: bug triaging imbalance multimodal fusion graph convolution |
