Design And Realization Of Question Answering System Based On Knowledge Graph

In recent years,with the development of the Internet and the expansion of knowledge,only a small part of the large amount of information on the Internet is useful to people.Search engines also have many deficiencies in dealing with this problem,such as too much noise,too little effective content,and inaccurate problem positioning,which results in users being unable to efficiently obtain effective information from the Internet.People are paying more and more attention to the question answering system because the question answering system can select effective information from massive information and feed it back to people.With the development of knowledge graphs,the model of traditional question answering systems has gradually changed.The recognition of the input question and the accuracy of the entity search based on the entities in the question are the key points to measure the quality of the question answering system.This paper uses question similarity calculation and improved TransE model to complete a question and answer system based on knowledge graph.The traditional knowledge representation learning model can effectively deal with one-to-one entity relationships,but it has poor accuracy for one-to-many or many-to-one entity relationships.In this paper,in-depth research on question matching,candidate entity acquisition,and knowledge representation model technology,using Chinese knowledge graph as the source of knowledge,improves the user's question entity recognition algorithm and traditional knowledge representation algorithm,and gives the innovation of this article.Algorithms and models were tested in real data,and a question and answer system based on knowledge graph was constructed.The main work of this article is as follows:(1)The accuracy of question understanding in the question answering system based on the knowledge graph is still insufficient.In this paper,the question matching step is added before the question entity recognition.First,word embedding is used to input Questions are embedded in words,the vectorized questions are compared with the set of constructed candidate questions,the optimal question is selected,and the entity of the selected optimal candidate question is identified,thereby improving the accuracy of question entity recognition.(2)In the traditional knowledge graph question answering system,after the entity search,the disordered entity link sequence is returned.In order to improve the accuracy ofthe target entity acquisition,this article will focus on the four aspects of the entity's popularity,semantic similarity,context similarity and character similarity.A multi-dimensional scoring mechanism is used to reduce redundancy,and an improved Rank M algorithm is proposed to rank candidate entities.(3)In view of the traditional knowledge graph representation model,TransE has a high degree of similarity to the representation of some entities,and it cannot represent one-to-many,many-to-one entity pairs.Joining the TransE model,an improved Tran EM model is proposed,which distinguishes the representation of highly similar entities and is tested by the data set.The experimental results prove that the method in this paper is improved compared with the traditional method.Based on the above algorithm and some DBpedia data sets,this paper implements a question and answer system based on Chinese knowledge maps,and conducts system tests in a real environment.The accuracy of the test results is higher than that of traditional question and answer systems.