Research And Application Of Key Technologies Of Limited Domain Question Answering System Based On Knowledge Graph

Question answering system is a research hotspot in natural language processing,information retrieval and other fields in recent years.Compared with the traditional search engine can only get complicated and vague information,question answering system can return accurate answers to users.Moreover,at present,knowledge mapping is also developing rapidly.Knowledge graph can organize fragmented knowledge through triples,establish the connection between data,and form a network,which is more conducive to information search,mining and analysis.Therefore,more and more people begin to research knowledge based question answering(KBQA).However,there are some difficulties in the research of question answering system which is based on knowledge graph.On the one hand,the current question answering methods are difficult to effectively complete the question answering system of complex questions involving multiple entities;on the other hand,the accuracy of traditional question answering method in attribute mapping is not high.The research of question answering system can be divided into open domain question answering system and restricted domain question answering system.Restricted domain question answering system is easier to put into use because of its small domain,obvious characteristics and higher algorithm accuracy,which is of great significance to improve the intelligent service level of professional field.This paper selects the college entrance examination consulting field to research question answering system,because the number of college entrance examination in China is large,there is a huge demand for information query when applying for the voluntary examination.The traditional way of information query is online search,book query,manual consultation,etc.,which is time-consuming and labor-consuming.If we can establish a question answering system that includes all the relevant information of the school,and return the answers quickly and accurately,then we can make up for the lack of information that is difficult to find when applying for school.In this paper,the key technologies of the restricted domain question answering system based on knowledge graph are studied in the field of college entrance examination consultation.The main contents are as follows:Firstly,the construction method of question classification system in KBQA field and the corresponding classification method are designed.The data model of university information knowledge graph is designed firstly according to the crawled questions,and then the question classification system is combined with the data model of knowledge graph.According to the entities and attributes in the data model,the question categories are divided into large and small categories,and the question classification is carried out in two steps.Secondly,the BERT-Bi LSTM model is proposed to improve the accuracy of similarity calculation,which not only takes advantage of the dynamic generation of word vector by combining the best with context,but also uses Bi LSTM to capture sentence information to improve the accuracy of similarity calculation.This model is used in the process of question classification and question answering.Thirdly,the paper studies the question answering method for complex questions,and proposes an improved question answering model based on semantic analysis.The query graph is generated by adding constraints,and then the query graph is automatically converted into Neo4 j query statements,and the answers are retrieved in the database.Fourthly,a knowledge graph of university information is built by the top-down way.According to the data model designed in the question classification stage,the information of colleges and universities is crawled from multiple websites,and then the format is sorted out and imported into neo4 j database.Then on the basis of this knowledge graph,a question answering system about the college entrance examination consultation is implemented based on semantic analysis.