Research And Application On Knowledge Graph Based On Question Answering System For Stroke

The intelligent question answering system is to sort out the disordered corpus information in an orderly and scientific way,and to establish a classification model based on knowledge for the human-computer interaction question answering.The application of intelligent question answering system in the medical field can provide medical treatment and health consultation for patients,so as to effectively relieve the strain of medical resources.In view of the problem that medical knowledge involves a large number of professional words,which are not well understood by ordinary patients,knowledge graph is an effective method to intuitively express the multidimensional association between knowledge.Therefore,the research and application of question answering system based on knowledge graph has important academic significance and application value in the medical field.Aiming at the problems of relatively specific and limited relational categories of medical question answering corpus information,this thesis proposes a deep learning model based on multi-granularity fusion to effectively classify the expected information of question answering.Furthermore,the knowledge base of stroke question answering was further constructed,and a prototype system of stroke question answering based on knowledge graph was designed and implemented.The main contents of this thesis include:(1)In the question answering system based on knowledge graph,entity recognition and relation extraction of questions are very important.Aiming at the problem that relations in medical questions can usually be covered by fixed categories,this thesis transforms the question of relation extraction into a multi-category problem.A Multi-granularity fusion neural network model(Mgf-Network)is proposed.It contains three core modules:Sequence encoding,Phrase vector recombination,and Semantic feature extraction.Give a question,each word in the question is encoded by BERT pre-training model in Sequence encoding module to obtain the word granularity vector.Phrase vector recombination module first performs question segmentation using"Jieba" word segmentation algorithm.Then it calculates each phrase granularity vector based on word granularity vector obtained by the Sequence encoding module.Finally,Feature extraction module extracts Multi-granularity fusion features from the first two modules.(2)This thesis designs and implements a stroke knowledge question answering system based on knowledge graph.First of all,this thesis acquires stroke related knowledge from medical websites and medical literature to build a stroke Q&A knowledge base,which concludes 2440 entities and 9767 triples.Then,they are stored in the gStore database.Secondly,this thesis uses the Bert+Bi-LSTM+CRF model and the Mgf-Network model proposed to perform entity recognition and relation extraction on questions.Then the extracted entities and relationships are mapped to SPARQL query statements to query the answer in the gStore database.Finally,the thesis presents the prototype realization of the system.