Research On Entity Matching Methods Based On Entity Context Representation Learning

In recent years,the scale of knowledge graphs and the number of entities has grown rapidly.The importance of entity matching between different knowledge graphs has become increasingly evident.The quality of entity matching depends on the contexts of the entity in the knowledge graph,and is mainly divided into three types: relationship triples,attribute triples,and entity text description.Related work doesn't model the three types of entity contexts at the same time.In addition,although some related work model two types of entity contexts,most of them doesn't treat these entity contexts equally,and the joint learning of multiple insufficient entity contexts cannot be solved well.In order to solve the above problems,this paper proposes a new entity matching method based on entity context representation learning.This method simultaneously models relation triples,attribute triples,and entity text description,because these entity contexts are crucial information for entity matching tasks.Then,the method considers the problem of insufficient information for a single type of entity context,and proposes three joint learning models.Based on the above,this paper first models the three types of entity contexts,then uses the three types of entity contexts to learn the joint embedding representation of the entities,and finally finds all matching entities by calculating the similarity of the entity embeddings.The main contributions of this paper are as follows:1)Proposing an entity matching model for modeling relationship triples,attribute triples,and entity text description,which uses efficient knowledge graph translation method to model relationship triples,graph convolutional network to model attribute triples,and recurrent neural network that is good at processing sequence data to model entity text description.2)Proposing three joint learning models that consider the problem of insufficient information for a single type of entity context.Based on the entity matching models of the three types of entity contexts,three joint learning models are designed,which are concatenate,entity context correlation analysis,and multi-view intact space learning.The joint learning models learn the joint embedding representation of entities,and then find the matching entities by entity embeddings.3)Performing experiments on cross-lingual datasets,and evaluating the model with widely-used criteria.The experimental results show that the proposed method outperforms the state-of-the-art method in most evaluation criteria.Researching knowledge graph entity matching method that merges different KGs into a unified KG,which not only has important application value in the fields of information retrieval,machine reading and question answering,etc.,but also have important theoretical significance for the development of Linked Open Data project.