| In recent years,with the Internet technology rapidly developing,a lot of professional knowledge has been saved into the Internet in various forms.How to obtain effective information from it has become increasingly challenging.Question answering system can resolve users’ questions through semantic information and make more concise answers,which gradually becomes a new way of information retrieval.In this thesis,taking the medical domain as research direction,we discuss the relevant method of building the domain question answering system.And we conduct indepth analysis and research on the technology of knowledge extraction and semantic parsing,improve the deep learning model.Finally,we build a medical question answering system based on knowledge graph.The main research work of this thesis includes the following aspects:(1)A joint extraction model RBLGP based on Ro BERTA-wwm and multi-layer Global Pointer is proposed to extract effective information from domain data into knowledge.This model choose the method of joint extraction of entities and relations and give up the traditional pipeline extraction method,which reducing the error accumulation and improving the information utilization;We introduce Ro BERTa-wwm pre-training model and Bi LSTM to extract text features,which improves the overall language understanding ability of the model;Multilayer Global Pointer are used to complete the task of marking triples,which effectively avoids the impact of overlapping entities and triples.(2)The semantic parsing model in the intelligent question answering method is improved.For entity recognition,a Ro BERTA-wwm-Bi GRU-Global Pointer model RBGGP is proposed;For text classification,RTT based on Ro BERTA-wwm-Text RCNNText CNN model is proposed;At the same time,based on the improved semantic parsing model,an intelligent question answering method based on semantic parsing is designed and implemented.(3)A domain question answering system based on knowledge graph is constructed.Extracting knowledge from internet websites by crawler to build a knowledge graph,using the joint extraction model to obtain the knowledge in the domain literature to supplement the graph,and then calculating the similarity of text and semantics to realize knowledge fusion,using Neo4 j database to store knowledge and build a domain knowledge map;Under the support of knowledge atlas,an improved intelligent question answering method based on semantic analysis is used to build domain question answering service;Finally,we integrate the front-end and back-end separate architecture and the microservice architecture,build the front-end and back-end respectively based on Vue and Spring Boot,and build the medical question answering system. |
