Research And Implementation Of People Relationship Question Answering System Based On Knowledge Graph Completion

With the rapid development of Internet industry and automatic question answering technology,there are more and more network information resources to be researched and developed.On the one hand,because users prefer simple and accurate search engines,the traditional retrieval method of keyword matching in natural language field is gradually changing to the direction of intelligent question answering.In the huge Chinese encyclopedia network information resources,the person relationship information is also a part of the digital industry.In order to better integrate the Internet resources into the intelligent question answering system,and make the generated products make the user experience better,this paper makes a preliminary research and exploration.On the other hand,in the task of knowledge map completion,it is essentially the exploration of predicting the unknown links in the fact,and predicting the missing parts in the triple through reasoning learning.In this paper,the method of knowledge representation learning is used to improve the existing model to compare the prediction effect.The method is applied to the query and retrieval of question answering system for reasoning completion.A person relationship question answering system based on knowledge map completion is designed and implemented.The main contents are as followsFirstly,in order to make up for the lack of character relationship resources,a certain scale of character relationship knowledge map is constructed.Using CBDB,Baidu Encyclopedia and Interactive Encyclopedia as data sources,the dataset collects and processes the data in two forms: mapping and extraction.It integrates 61318 Chinese character entities,128 people relationship types and 151892 facts.Secondly,aiming at the problem of the lack of the content of the three tuples in the knowledge map,this paper studies the knowledge map completion task in the question answering system,improves the traditional knowledge map embedding model TransH,and proposes the hyperplane projection model DTransH based on the flexible translation principle,It is proved that the distribution of the improved model in the vector representation of entities and relationships is more reasonable,and it also shows its effectiveness.In order to better transform natural language into semantic representation that can be understood and executed by machine and apply it in person relation question answering system,this paper uses DTransH method to complete the knowledge graph.Finally,on the basis of the completion experiment,the character relationship question answering system is designed and implemented.In the knowledge map person relationship question answering system,this paper preliminarily studies the problem that the system can deal with the retrieval of missing information,and proves that the reasoning completion of knowledge map can improve the retrieval accuracy and question answering matching accuracy in the person relationship question answering system.The existing technical means are used to process the data of person relationship structurally and apply it in question answering system.The stored person relationship triples are used for reasoning and prediction in question answering,and the graph database is used to store the knowledge map of person relationship,which is convenient for the answer query and knowledge base retrieval of person relationship question answering system.In view of the natural language questions inquired by users,the system uses the question preprocessing module to analyze the question sentences,generates the question triples,generates the query sentences through the answer generation module,searches and obtains the answers in the graphic database,and displays them visually in the person relationship question answering system.