基于属性增强和关系感知的图卷积实体对齐方法  

Entity alignment based on attribute enhancement and relationship awareness

作　　者：高兵 黄超 邹启杰[1,2] 秦静[1,2,3] GAO Bing;HUANG Chao;ZOU Qi-jie;QIN Jing(College of Information Engineering,Dalian University,Dalian 116622,China;Key Laboratory of Intelligent Medicine and Health of Dalian,Dalian University,Dalian 116622,China;School of Software Engineering,Dalian University,Dalian 116622,China)

机构地区：[1]大连大学信息工程学院,辽宁大连116622 [2]大连大学大连市智慧医疗与健康重点实验室,辽宁大连116622 [3]大连大学软件学院,辽宁大连116622

出　　处：《计算机工程与设计》2024年第5期1384-1390,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62002038);辽宁省科学研究经费基金项目(LJKZ1180)。

摘　　要：实体对齐是知识融合最重要的步骤之一,在知识图谱构建和融合过程中,往往存在结构不相似、实体表述不够准确,甚至知识缺失的问题。针对以上问题,提出一种基于属性增强和关系感知的图卷积实体对齐方法,结合实体和关系之间的内在联系,通过实体可以推导出关系,通过关系可以表示实体,利用实体的属性增强实体表示,增强实体对齐的效果,设计一种迭代策略,迭代增强关系和实体结合后的实体对齐效果。在通用数据集上进行的大量实验结果表明,相对于原有基于实体嵌入的方法,所提方法具有更高的有效性和对齐率。Entity alignment is one of the most important steps of knowledge fusion.In the process of knowledge map construction and fusion,there are often problems such as different structures,inaccurate entity representation,and even the lack of know-ledge.In view of the above problems,a graph convolution entity alignment method based on attribute enhancement and relationship awareness was proposed.Combining the internal relationship between entities and relationships,relationships were derived from entities,and entities were represented through relationships.The entity representation was enhanced using the attributes of entities,and the effect of entity alignment was then enhanced.An iterative strategy was designed to iteratively enhance the effect of entity alignment after the combination of relationships and entities.Results of a large number of experiments on general data sets show that the proposed method has higher efficiency and alignment rate than the original entity embedding method.

关 键 词：实体对齐 知识融合 知识图谱 属性增强 关系感知 图卷积 迭代策略 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]