Health Knowledge Question Answering System Based On Knowledge Graph And Deep Neural Network

With the rapid development of technology and the economy,search engines have taken the lead in providing a way for users to find medical and health information.However,as the volume of information has increased geometrically,the problems that search engines have shown in terms of questions and answers have become increasingly apparent.Problems such as inaccurate answers,cumbersome answers,inability to accurately understand user requests,and inefficient feedback.The rapid development of the Knowledge Graph has provided data to support various other areas.Knowledge graphs preserve entities and related attributes in the form of triples and describe realworld entities and relationships in the form of semantic networks.By combining health knowledge knowledge graphs with question and answer systems,answers can be provided directly and concisely.It also speeds up the dissemination of health knowledge.In this thesis,we propose the named entity recognition model BERT-IDCNN-CRF,and based on this model,we design and implement a knowledge graph-based health knowledge Q&A system,which effectively improves the recognition rate of questions and the correct rate of question responses.The main research contents and contributions of this thesis are.1.This thesis constructs a knowledge graph in the field of health care: using Python crawler technology to crawl medical and health-related data from "Family Doctor","Seeking Medicine" and "39Health.com" respectively.Crawl health-related data from "Family Doctor","Find Medicine" and "39 Health".Knowledge extraction and knowledge fusion are carried out on the data to obtain entities and relationships related to the healthcare domain,and finally the healthcare knowledge graph is constructed and stored in the Neo4 j database.2.In this thesis,a BERT-IDCNN-CRF named entity recognition model is built:considering that Bi LSTM cannot use GPU parallel computing,resulting in low training efficiency of the model,IDCNN is chosen instead of Bi LSTM in this thesis,which can be regarded as a CNN after adding the expansion width.In order to obtain a broader input matrix,the CNN can be considered as a CNN with the addition of swelling width,which can ignore the data between the swelling widths when the convolution kernel slides in the continuous region.At the same time,in order to prevent the emergence of illegal label sequences,entity recognition is accomplished in combination with CRF.Through multiple sets of comparison experiments,it is demonstrated that the model in this thesis can effectively complete the named entity recognition in health knowledge.3.In this thesis,a health knowledge Q&A system based on knowledge graph and deep neural network is constructed.Based on the constructed medical and health knowledge graph and the designed named entity recognition model,a health knowledge quiz system based on the knowledge graph is implemented.The system design and the technical route of the system are analysed first.The system is divided into three main parts: question parsing,template matching and query execution.In the question parsing part,the BERT-IDCNN-CRF model extracts the entities in the questions,extracts the keywords in the questions using the word separation tool,and matches them with the preset templates.The final query statement is generated based on the entities as well as the user intent to complete the retrieval of the answer.The system can provide an interface to interact with the user in real time,giving the results of the question and answer in the form of text feedback to the user,as well as recording the questions that cannot be answered.The experimental results show that the application of the named entity recognition model proposed in this thesis to a health knowledge quiz system can effectively improve the effectiveness of the quiz and facilitate the dissemination of medical and health knowledge.