Research On Entity Alignment Method Of Cross-language Knowledge Graph Based On TransD

In recent years,with the rapid development of the Internet,more and more knowledge graphs appear,which may have the problem of content duplication.A single knowledge graph is a representation of knowledge organization,which is used to describe a variety of concepts,entities or attributes in the real world.However,researchers also realize that a single knowledge graph cannot meet people’s needs,so researchers set out to study entity alignment across languages,which can effectively integrate different knowledge graphs,expand information content,and reduce redundancy.Finally,a large knowledge graph with clear logical structure and complete content is formed.The difficulty of entity alignment is how to determine the equivalence of entities in two different forms of knowledge base.Cross-language entity alignment method based on translation model and cross-language entity alignment method based on graph neural network model are existing entity alignment methods.On the one hand,the entity alignment method described by most literatures is based on the Trans E model,which is not suitable for complex relationships,such as "one-to-many" or "many-toone",but there will be one-to-many,many-to-one and many-to-many complex relationships between entities,and there is no model to establish multiple relationships across languages.They also ignore the multistep relational path information.On the other hand,entities have diversity and relationships have intrinsic relevance.In addition,traditional methods ignore the influence of search speed,and most iterative processes calculate entity similarity in one direction by default,which will introduce wrong target entities in the process of entity alignment,leading to propagation errors and other problems.In order to solve the above problems,the research content of this paper is as follows:(1)To solve the problem of ignoring multiple complex relationships and multistep path information between entities,this paper proposes a TransD-based crosslanguage entity alignment method(TransDEA).Since TransD has traditionally been used only in a single knowledge graph,this article is the first attempt to use the TransD approach to handle entity alignment across languages.Due to the existence of multistep path information between entities in KG,the paper tries to improve the TransD method by using the multi-step path information,namely the relationship between second-order neighbors of entities,to strengthen the establishment of complex multivariate relationships.On this basis,the paper adds parameter sharing and bootstrapping strategies to deal with relational triples,which can better deal with the above problems.Parameter sharing is the representation of previously aligned entities as the same vector when the projection is embedded.Experimental results on real world data sets show that this method is better than the benchmark method.Experiments show that the proposed model is helpful to improve entity alignment.(2)Cross-language entity alignment using TransD model proves to be more efficient than the traditional approach using Trans E model.In view of the diversity of entities and the inherent relevance of relationships,this paper adds edge embedding on the basis of TransD.In order to better improve the search speed,two-way iteration strategy and re-initialization process are added on the basis of TransDEA model to expand entity seeds.Therefore,a bidirectional iterative entity alignment method(Bi Tr DCP-Align)combining TransD and edge embedding is proposed in this paper.The bidirectional iterative strategy is used to calculate the equivalent entity of each entity.It is a semi-supervised learning method.The aligned seeds are already available before entity alignment.On this basis,new equivalent entity pairs are calculated bidirectional,and then added into the entity alignment seed set.Bi Tr DCP-Align’s Hits@k score on DBP15 K data set is significantly improved,which proves that adding edge embedding and bidirectional iteration strategies and reinitialization strategies can improve the search speed and entity alignment effectively,and reduce the occurrence of errors.