Design And Implementation Of Medical Intelligent Question Answering System Based On Knowledge Base

Objective:In recent years,with the rise of online medical service,a large number of online medical and health platforms have emerged.Patients describe their symptoms,problems on the platform and the doctors make diagnosis,advice,or further inquiries according to the description of the patients.This can solve the problems of some patients remotely,without going to the hospital which greatly facilitates the patients,and also plays a certain role in the rational allocation of medical resource.But with the accumulation of data,many patients put forward similar problems,resulting in a waste of medical resources.Therefore,based on a variety of natural language processing technology,and relying on the publicly available medical data such as medical question answering database,we build an intelligent question answering system to serve patients.Methods:Firstly,the medical domain question answering knowledge base and drug-related knowledge graph were constructed,and the synonymous corpus preparation platform was designed and built.Taking the Chinese medical questions in the knowledge base as the template,the medical similar question data set was prepared.Then,the corpus was used to train the medical similar question matching model,which would be used to compare the patient's question with all the questions in the question answering knowledge base to find the most similar question and return the corresponding answer as the response.For the questions related to specific drugs raised by patients,the answers will be given based on the drug knowledge base.Finally,the depth matching model of medical similar questions and the drug question answering model were integrated to build the medical intelligent question answering system.Results:We used two public synonymous text databases to verify the synonymous corpus preparation platform.The results show that the similar text matching model trained with the data prepared by this platform and the model trained with the original data perform equally on the test set,indicating that it is feasible to prepare synonymous corpus with this platform.Limited by the size of the question answering database,the intelligent question answering system built in this paper can answer about 70%of the patients' questions,and the acceptability of the answers is P@1=0.47,P@3=0.6,which can make a good answer to the common questions raised by the patients.Conclusion:With the support of medical question answering knowledge base,the method of similar question matching can answer about 69%of patients' questions.When patients' questions involve personalized drugs,the effect of question answering can be further improved by accessing knowledge graph.The results show that with the support of massive question answering data and medical knowledge database,the natural language processing technology can automatically answer patients' common questions,which can reduce the burden of doctors to a certain extent.