Research And Application Of Auxiliary Diagnosis System Based On Medical Knowledge Graph

With the vigorous development of our society and the gradual improvement of network infrastructure construction,people can more easily access the Internet through personal devices such as computers and mobile phones,and enjoy the various conveniences brought to our lives by the wave of informatization.When people need help with medical and health problems,in addition to going to the hospital to see a doctor and following the doctor’s advice,they can now use the almost massive amount of medical knowledge on the Internet to complete self-diagnosis.However,while the rich information on the Internet is convenient for people,it also brings about the problem of information overload.When people search for information about a certain disease,search engines often return nearly tens of millions of different search results based on keywords.People need to search for the information they really need in so many search results.This is not a small burden for most people.In response to the above problems,this thesis attempts to build a medical questionand-answer system based on the knowledge graph to help users with auxiliary diagnosis.Different from the search results that search engines directly return keywords,the medical question-and-answer system will further analyze the user’s questions,and then only return the answer that is closest to the user’s search intent.For users,it saves complicated information screening work and reduces the difficulty of obtaining medical information.The main work of this thesis consists of 4 parts:(1)The overall structure of an auxiliary diagnosis system based on medical knowledge graph is proposed,and the system is divided into six main modules:knowledge graph building module,named entity recognition module,entity linking module,intention recognition module,answer query module and answer construction and display module.(2)The knowledge graph module first crawls the disease encyclopedia data of the medical and health website,and then constructs a medical knowledge graph containing about 40,000 entities and about 200,000 entity relationships.(3)For the medical named entity recognition module,the BERT-based multi-level convolution-CRF network proposed in this thesis has a comprehensive F1 value of 0.886,which significantly exceeds the baseline model;for the entity linking module,an efficient algorithm is designed to point the similar described entity to the knowledge graph entity;for the medical intent recognition module,the accuracy of the BERT-based intent classification model proposed in this thesis has reached 92.83%,which is 3.71% higher than the basic Text CNN.(4)Finally,the Flask lightweight development framework is used to integrate the various modules of the system,add an easy-to-use user interface,and complete the construction of the entire auxiliary diagnosis system based on the medical knowledge graph.This thesis aims to solve the problem that people are currently encountering in the field of medical and health that it is difficult to filter search information,and through deep learning models and natural language processing methods to improve the semantic understanding of the entire auxiliary diagnosis system,and finally build a simple and easy-to-use Auxiliary diagnosis system.