基于交叉图匹配和双向自适应迭代的实体对齐

Entity Alignment Based on Cross Graph Matching and Bidirectional Adaptive Iteration

实体对齐(Entity Alignment,EA)是在多源知识图谱中寻找更多等价实体对,是构建知识图谱过程中极其重要的研究任务,而目前知识图谱间的异构和标记种子数量不足是主要问题。基于此,提出一种基于交叉图匹配和双向自适应迭代的实体对齐方法,使用交叉图信息增强图间关系的交互,双向自适应种子迭代的增加训练种子数量和质量。在3个真实数据集的基础上进行验证,有效提高了实体对齐效果。

Entity Alignment(EA) aims to discover more equivalent entity pairs between knowledge graphs, which is an extremely important research task in the process of constructing knowledge graphs. At present, the heterogeneity of knowledge graphs and the shortage of marker seeds are the main problems. This paper proposes an entity alignment method based on cross graph matching and bidirectional adaptive iteration. The cross graph information is used to enhance the interaction between graphs, and the bidirectional adaptive seed iteration increases the number and quality of training seeds. Verified on three real datasets, it can effectively improve the entity alignment effect.