Research On The Method Of Link Prediction Between Entities Based On Knowledge Graph

The development of knowledge graph technology provides an effective solution for the industry and academia to better organize,manage and understand the massive amounts of data on the Internet.At present,knowledge graph technology is widely used in many AI fields such as entity search,recommendation system,and open domain question and answer.However,there are missing entities or links in the knowledge graph,and the knowledge graph is not complete which greatly limits the accuracy of the knowledge graph for retrieval and reasoning.Knowledge graph completion which is to complete the task of predicting missing links has become one of the core tasks of knowledge graph research.This thesis conducts a comprehensive investigation and introduction to the link prediction algorithms of the knowledge graph,and analyzes the graph embedding models based on translation,the reasoning based on logic rules,the graph embedding based on tensor,and the graph embedding model based on deep learning network.Two improved models are proposed for the existing problems of the current graph embedding models,and they are applied to the link prediction system.The main research contents of this thesis are as follows:1)This thesis analyzes the problems of the translation-based knowledge graph embedding models,and proposes a multi-semantic entity and relationship embedding model.In the models' hypothesis,the magnitude of the vector is ignored,and the parallel relationship is retained,so that the entity or relationship vector shows the same direction and different magnitude in different triples,so as to solve the multi-semantic problem of entities and relationships.2)The current mainstream graph embedding models ignore the structure information of the graph,and only consider the information in a triplet.The proposal of the graph attention network provides an effective means for extracting the structural characteristics of the graph.However,the graph attention network only considers the influence of the entity node on the selected entity,and ignores the influence of the relationship between the two entities.Accordingly,this thesis proposes a multi-hop embedding model that introduces an attention mechanism.3)In order to complete the task of completing the knowledge graph,this thesis designs and implements a link prediction system based on the knowledge graph embedding models.Using the ternary data set of the knowledge graph to train the model,the input problem is predicted and the proposition is judged.Provides a system interface for adding and deleting knowledge graph triples,so as to improve and complete the knowledge graph.