Research On Medical Knowledge Graph Question Answering System Based On CNN Model

Since the outbreak of virulent infectious diseases such as COVID-19,the lack of medical resources has become a serious social problem,and online Q&A systems in the medical and health field have become one of the important ways for people to obtain health knowledge and treatment advice.However,due to the professionalism and complexity of the medical field,the lack of accuracy of medical Q&A systems has become one of the key factors limiting their development.The development of technologies such as big data,artificial intelligence,and knowledge graphs has provided new ideas and methods for medical Q&A systems.As an emerging knowledge representation and reasoning technology,a knowledge graph can integrate professional knowledge and data in the medical field into a visual knowledge graph to establish an accurate,comprehensive,and scalable knowledge system to support the construction and optimization of medical Q&A systems.This paper accomplishes the following research work:Firstly,the medical knowledge graph NWNU＿KG is constructed.Structured data are extracted from multiple data sources such as medical electronic documents,vertical medical sites and medical symptom repositories by establishing rules,and the data from medical sites are crawled using a crawler parser.After operations such as entity alignment and entity disambiguation,the acquired medical data are fused to construct the medical knowledge graph NWNU＿KG,which contains 7 types of 44112 entities and291164 relationships of 10 types.Secondly,the medical question-and-answer federated model MBCD is proposed based on the constructed knowledge graph.To solve the problem of entity recognition,the MBCD model uses a combination of deep learning and lexicon.The deep learning model MC-BERT＿BILSTM＿CRF recognition failure then uses the lexicon for secondary entity recognition to improve the accuracy and coverage of entity recognition.For question intent recognition,the MBCD model adopts a combination of text CNN and rule templates.Text CNN model fails to recognize user input questions and then matches possible user intents by invoking the set rule templates to achieve secondary intent recognition.The proposed MBCD model effectively solves the problem of entity extraction and question intent recognition in medical Q&A systems,and further improves the accuracy and performance of Q&A systems.Finally,a medical Q&A system based on the knowledge graph NWNU＿KG and MBCD model is developed.The system is built with a data layer,business layer,and display layer,and the front-end display page is developed by HTML and CSS technology and built quickly by the Django framework.It mainly implements entity relationship query and Q&A interaction and knowledge graph visualization and can provide medical Q&A service and medical knowledge graph visualization applications and other functions.This thesis finds that the medical knowledge map NWNU＿KG is scalable and easy to migrate,which can provide data support for subsequent intelligent Q&A research in the medical field and has broad application prospects;the joint model MBCD has a high accuracy rate,with 89.32% and 85.68% in entity recognition and relationship extraction,respectively,and can accurately identify the natural language of the user input.The MBCD model-based intelligent medical diagnosis and treatment service platform has fast response speed,a simple and easy-to-understand human-computer interaction page,and can accurately identify user questions and return correct answers,with good stability and reliability.