基于图神经网络的实体对齐研究综述

Review of Knowledge Graph

实体对齐是知识图谱构建中的一项重要任务,也是近年来研究热点之一。通过实体对齐,可以将互相隔离的,不同来源的知识图谱汇总聚合,构成包含信息更为丰富新知识库。但是实体对齐任务往往面临诸多问题,如不同来源的知识图谱的语言,关系类型与结构都有可能不同,这也就使得很难通过简单的翻译与变换来进行对齐任务。近期涌现较多基于图神经网络的实体对齐方法,有效地克服之前的缺陷,对这些方法进行总结与汇总,并介绍常用的相关数据集。

Entity alignment is a key task of knowledge graph building,and also is a hot topic in recent years.We could link and fusion different knowl edge graph into a new knowledge graph with more information by entity alignment method.But there are many difficulties in these task,for example,different knowledge graph from different sources usually don’t have same structure or same link types,and these different will make it hard to match entities by only using translation.Most recently,several entity matching method based on Graph Neural Networks(GNN)has been proposed for this task.This paper reviews and summarizes these new method above and introduces the commonly used da taset in alignment.