Automatic Question Answering System Based On Legal Knowledge Graph

With the gradual perfection of China’s characteristics socialist rule of law system,people’s legal awareness is rising and the demand for legal services is growing.However,lawyers in China are still scarce resources.The number of lawyers cannot match such a huge market of legal consultation,and legal consultancy service cannot satisfy the growing market.From the aspect of users,legal consultation is costly.it is difficult to obtain professional,timely and free legal consultancy services.From the perspective of lawyers,the legal issues are trivial and complicated,and attorneys specialize in different areas,resulting in the uneven quality of responses,lack of motivation to answer questions due to a mismatch between pay and return.To achieve the goal of“digital rule of law,smart justice”,it is necessary to provide fast,free and accurate legal consultation services、improve the quality of legal services、reduce the cost of user consultation and lawyer service、alleviate the contradiction between the supply and demand of legal advisory services.Therefore,it is of great application value and practical significance to study automatic legal question and answering(Q&A)methods.In this thesis,an automatic question-answering system based on a legal knowledge graph is designed and implemented to address the shortcomings of the existing legal question-answering system,such as difficulty locating answers quickly,inability to reason,and limited by the question-answering library.The system is mainly divided into legal knowledge graph construction module,legal knowledge Q&A module,legal knowledge retrieval module and legal entity recognition module.At present,the legal question and answer model does not make sufficient use of legal knowledge and manual response is not professional.To solve the problems,this thesis constructs a knowledge graph in the legal field,based on the legal data of ”National Database Of Laws And Regulations”,supplemented by the data of third-party legal service websites and Baidu Encyclopedia,through ontology construction、knowledge acquisition、knowledge extraction and knowledge fusion.To automatically identify legal entities,help computers quickly understand questions and extract legal knowledge topics,provide technical support for other modules,this thesis constructs a legal entity recognition model based on ALBERT+Bi LSTM+CRF.An entity recognition corpus is constructed using the BIO annotation method to train the legal entity recognition model.The experimental results show that the model has excellent results in the legal entity recognition domain.legal knowledge Q&A module realizes the standardization,automation and intelligence of the legal consulting service.For the input legal questions,the legal entity recognition identifies the relevant legal terms,Fast Text model extracts the labels of complex questions,and the legal knowledge retrieval module returns a series of candidate answers triples.To select the final answer from candidate answer triples,the pre-training model ALBERT is used to extract the attributes of the question。In this thesis,we provide an efficient,professional and zero-threshold legal knowledge Q&A system for users with a web application based on Django framework.Users can access this system through a browser to ask a legal question,query legal knowledge and query legal relationship.The system performs the corresponding task according to the input text and displays the result.The system uses a knowledge graph to associate legal domain knowledge with the input text and displays them in the browser by EChart frame.After experiments,for standard legal questions,the correct rate of response reaches 86.08%;for complex legal questions,the correct rate reaches 69.04%,and the corresponding speed of question answering is less than 1 second,which can answer user questions quickly and relatively accurately.