实体对齐技术是知识图谱研究中的重要方向,旨在将不同的知识图谱中指向同一现实对象的不同实体进行连接,进而实现知识图谱的扩充.目前,该领域的主流研究思路有2种:一是针对知识图谱的结构特征进行分析,二是针对知识图谱的元素信息(即实体名、关系名、属性名等)进行分析,但是没有同时针对结构特征与元素信息同时进行分析的模型.提出一种实体对齐模型EAFF(entity alignment based on feature fusion),该模型从图结构与元素信息2个角度,对知识图谱进行特征分析.首先,设计了一种基于图神经网络的实体对齐算法,以获得基于图结构的对齐实体对;其次,设计了一种基于元素信息的实体对齐算法,以获得基于元素信息的对齐实体对;最后,利用特征转换与排序算法对2组对齐实体对进行排序,进而得到知识图谱中的对齐实体.在实验中,EAFF取得了相对较好的成绩,优于目前的主流算法.
Entity alignment based on graph structure and element information fusion
Entity alignment,as an important research direction in knowledge graph research,aims to connect different entities pointing to the same real-world object in different knowledge graphs,and thus to achieve the expansion of knowledge graphs.At present,there are two mainstream research approaches in this field.One is to analyze the structural characteristics of knowledge graphs,and the other is to analyze the element information(such as entity name,relation name,attribute name)of knowledge graphs.In this article,a novel entity alignment model EAFF(Entity Alignment based on Feature Fusion)is proposed to analyze the features of knowledge graphs from the perspectives of graph structure and element information.First,a graph neural network-based entity alignment algorithm was designed to obtain aligned entity pairs based on graph structures.Then,an entity alignment algorithm based on element information was designed to obtain aligned entity pairs based on element information.Finally,using feature transformation and sorting algorithms,two sets of aligned entity pairs are sorted to obtain aligned entities in the knowledge graph.In the experiment,EAFF achieved relatively good results,surpassing current mainstream algorithms.
NETL
NSTL
万方数据
