Research On Entity Profiling Method In Knowledge Graph

Knowledge Graphs are graph-structured knowledge bases storing factual information about real-world entities.Recent years with a rapid growth in knowledge graph constructions,each entity description in knowledge graph was associated with more and more facts,which is unacceptably long when users want to quickly compare and understand entities.The research purpose of entity profiling in knowledge graph is finding the most distinguishable structural features to profile the entities and help the users identify or compare the entities efficiently.State of art,there is no specific research about the entity profiling problem in knowledge graph,so an entity profiling method which is based on structured-labels is proposed.The result of entity profiling can simplify the overloaded description of entities.The main research contents of the thesis are as follows:(1).The method for constructing structured-labels of entities in knowledge graph has been proposed.Based on the analysis of the complex entity description,statistical inference,heuristic rule filtering,structured-labels ranking techniques are used to construct the structured-labels automatically.(2).The method for measuring the distinctiveness of structured-labels has been proposed.Based on the distinctiveness formula,candidate structured-labels are ranked and filtered according to the distinctiveness score.Three methods are proposed to measure the similarity of entities,including the use of naive SimRank score,Monte-Carlo based estimation SimRank score and entity embedding vector distance based on network representation learning to measure the similarity between entities in knowledge graph.Then the similarities of entities are used to measure the distinctiveness of structured-labels.(3).Design experiments to evaluate the feasibility of structured-labels construction method and the rationality of the entity profiling results.Finally,the entity profiling in knowledge graph is visually presented to users.The effectiveness of the proposed methods has been verified experimentally on real datasets.The entity profiling method based on structured-labels can mitigate the problem of entity identification and comparation in entity comprehension.The research of entity profiling in knowledge graph provides a new idea and method to promote entity comprehension,which has both theoretical and practical value.