| Recently,knowledge graph technology,an important part of artificial intelligence field,has attracted the attention of many researchers.A single knowledge graph is difficult to meet the actual application requirements,and there are problems such as redundancy and heterogeneity between different knowledge graphs.Therefore,knowledge graph fusion become a hot issue in this field.Entity alignment is one of the important tasks.It aims to discover entities from different knowledge graphs that represent the same thing in the real world.Most of the existing work focuses on structural information.However,there are still many kinds of feature information that have not been fully utilized.The usage of various feature information can improve the effect of knowledge graph embedding,making the embedding model more robust.This paper focuses on embedding multi-view knowledge graph,solving the problem of over-reliance on supervised data and using relationship or attribute alignment to enhance entity alignment.This paper proposes a novel entity alignment framework based on multi-view knowledge graph embedding,Multi KE.The framework not only can integrate and utilize multi-view information for entity alignment tasks,but also has good scalability.The main contributions of this paper are as follows:1.This paper defines three representative views,namely,name view,relation view and attribute view.It also develops an appropriate embedding model for each view.2.This paper proposes an entity alignment framework based on multi-view knowledge graph embedding and designs two cross knowledge graph training methods.3.To obtain the final entity representation for the entity alignment task,this paper proposes three view combination strategies to combine the multiple views embedding.4.The experiments show that Multi KE is significantly better than the existing related methods in entity alignment tasks.Moreover,this paper explores the performance of Multi KE in various aspects through several verification experiments. |
PDF Full Text Request