| 本文已被：浏览 2203次 下载 5606次 |
码上扫一扫！ |
|
|
| 面向列语义识别的共现属性交互模型构建与优化 |
|
高珊1,2, 袁宛竹1,2, 卢卫1,2, 王兰1,2, 张静1,2, 杜小勇1,2
|
|
1.数据工程与知识工程教育部重点实验室(中国人民大学), 北京 100872;2.中国人民大学 信息学院, 北京 100872
|
|
| 摘要: |
| 政务数据治理正在经历从“物理数据汇聚”到“逻辑语义汇通”的新阶段.逻辑语义汇通是指针对各孤岛政务系统因长期“自治”而形成的元数据缺失、元数据同名不同义以及同义不同名等问题，在不重建或修改原系统代码以及不物理汇聚各政务数据的前提下，通过技术手段，统一各孤岛信息系统元数据的语义表达，实现元数据的语义互联互通.该工作是将各孤岛信息系统的元数据语义对齐到已有的标准元数据上，具体地，将标准元数据名称看作语义标签，对孤岛关系数据的列投影进行语义识别，从而建立列名和标准元数据的语义对齐，实现孤岛元数据标准化治理.已有基于列投影的语义识别技术无法捕捉到关系数据的列顺序无关性特征以及属性语义标签之间的相关性特征，针对这一问题，提出了基于预测阶段和纠错阶段的两阶段模型：在预测阶段，提出了共现属性交互的CAI模型（co-occurrence-attribute-interaction model），利用并行化的自注意力机制保证列顺序无关的共现属性交互；在纠错阶段，结合语义标签之间的共现性，通过引入纠错机制（correction mechanism），优化CAI模型预测结果.在政务基准数据和Magellan等多组公开英文数据集上进行了实验，结果表明，引入纠错机制的两阶段模型，在宏平均和加权平均两个指标上，比已有最优模型最多可分别提高20.03%，13.36%. |
| 关键词: 孤岛政务 逻辑语义汇通 列语义识别 共现交互 注意力机制 |
| DOI：10.13328/j.cnki.jos.006787 |
| 分类号: |
| 基金项目:国家重点研发计划（2020YFB2104101） |
|
| Construction and Optimization of Co-occurrence-attribute-interaction Model for Column Semantic Recognition |
|
GAO Shan1,2, YUAN Wan-Zhu1,2, LU Wei1,2, WANG Lan1,2, ZHANG Jing1,2, DU Xiao-Yong1,2
|
|
1.Key Laboratory of Data Engineering and Knowledge Engineering (Renmin University of China), Ministry of Education, Beijing 100872, China;2.School of Information, Renmin University of China, Beijing 100872, China
|
| Abstract: |
| Government data governance is undergoing a new phase of transition from "physical data aggregation" to "logical semantic unification". Thus far, long-term "autonomy" of government information silos, lead to a wide spectrum of metadata curation issues, such as attributes with the same names but having different meanings, or attributes with different names but having the same meanings. Instead of either rebuilding/modifying legacy information systems or physically aggregating data from isolated information systems, logical semantic unification solves this problem by unifying the semantic expression of the metadata in government information silos and achieves the standardized metadata governance. This work semantically aligns the metadata of each government information silo to the existing standard metadata. Specifically, the standard metadata names are viewed as semantic labels, and the semantic meanings of columns of relations in each government information silo are semantically identified, so as to establish the semantic alignment of column names and standard metadata and achieve standardized governance of silo metadata. |
| Key words: government information silo logical semantic unification column semantic recognition co-occurrence interaction attention |
