Research On QA Model Based On Answer Selection Combined With Knowledge Graph

Question and Answering(QA)system can meet the needs of people who want to obtain information quickly and accurately.Although scholars have made great progress in the research of QA system,there are still some problems.Most of the current QA systems are based on the similarity between the question and the question,or the similarity between the question and the answer.When users' questions exceed the range of the system's training corpus,it will cause the decline of the accuracy of algorithms.In order to solve this problem,this paper applies the information of the knowledge graph to the QA model to expand the scope of the QA system.At the same time,in order to better use and learn the information of the knowledge base,this paper uses the known facts of the knowledge graph,combined with the common points of answer selection(AS)and knowledge base question and answer(KBQA),and adopts traditional QA model to improve knowledge base retrieval accuracy.The main content of this article includes the following points:(1)AS-KBQA model design: This article proposes the AS-KBQA question and answering model,which is divided into three modules: question understanding module,answer selection module and translation module.Question understanding module uses a combination of CNN and bi LSTM to extract the entity of a given question and the relationship between the question and the answer,and designs a heuristic algorithm for extracting entities through a lot of exploration.Answer selection module uses bi LSTM combined with Attention and Soft Max deep to select a set of candidate answer entities.Translation module includes three parts: 1.Use the BERT model to map the knowledge base information to the vector space,and convert the entity relationship into a space vector;2.Process the data passed by the two modules;3.Combine the trained Trans R model,retrieve the answer.(2)Construction of knowledge graph in the medical field: This article crawls a large amount of data from two large medical websites,and processes the collected data,including the processing operations of entities,data attributes,and relationships,and then constructs a Chinese knowledge graph in the medical field.For AS-KBQA model,two experiments are carried out in this paper to verify its accuracy and adaptability.Experiment 1: Select Freebase which provides a lot of open source data(English)as the knowledge base,conduct experiments on the three datasets of Simple Question,Yahoo QA and Web QSP,and select six QA models proposed in recent years as comparative experiments.With accuracy as the evaluation index for each model,the AS-KBQA model in this paper has better results on the two data sets of FB2 M and FB5 M.The experimental results prove that the answer selection model based on the knowledge graph can improve the accuracy of the knowledge base question answering.Experiment 2: Choose the medical domain knowledge graph(Chinese)constructed in this paper as the knowledge base,and conduct experiments on the MIE dataset to verify the adaptability of the AS-KBQA model.The experimental results show that the model can not only adapt to English QA,but also adapt to Chinese QA in different fields.Therefore,the AS-KBQA model proposed in this paper has high accuracy and strong adaptability.The application range of QA system can be extended only by extending the information in the knowledge graph of AS-KBQA model.