Research On Medical Automatic Question Answering With Pre-training Language Model

With the popularization of the Internet and the rapid development of mobile applications,people are getting more and more accustomed to getting information on the Internet.Traditional medical consultation requires people to go to the hospital to register for diagnosis.With the development of network technology,more and more hospitals have introduced online consultation services.People are gradually getting used to search engines or specialized question-and-answer sites,to search or ask questions about symptoms,and to obtain references from professional doctors’ answers.However,the resources of professional doctors are limited,and most of the medical problems are concentrated on common diseases such as colds and fevers.The answers to these common medical problems have great referential values.How to give a quick feedback to patient’s consultations by analyzing the existing medical data is a key issue.The proposal of the pre-training language model BERT refreshed the performance record of many tasks in the field of natural language processing.However,because BERT learns the feature representation on the basis of massive corpus,the effect in specific fields can be further improved,so it is of practical significance to try to adapt the field to continue pre-training methods.Aiming at the above problems,this paper tries to apply the pre-training language model to the automatic question answering system,uses domain corpus to continue pre-training the model,and proposes a BERT-CBOW model with extended sentence vector representation to improve the accuracy of question answering.The specific work and innovations of this paper are as follows:(1)A text similarity calculation method based on a pre-trained language model is proposed,and the retrieval-style automatic question answering is realized by this.Use the question and answer corpus in the medical field to continue pre-training the BERT model to further improve the accuracy of the model in the field question and answer.(2)Propose the BERT-CBOW model.Although the sentence vector represented by BERT can obtain a relatively high similarity value in similar sentences,the matching accuracy of similar sentences can be further improved.This paper uses the CBOW model to extend the sentence vector representation of BERT to make it a new sentence vector.The experimental results show that the top-k accuracy of the BERT-CBOW model in the test set of this article is improved compared with the baseline model and the model after continuing pre-training.(3)Design and develop an online medical automatic question and answer application system including server,web and mobile terminals to solve actual question and answer questions in the medical field and provide relevant references for questioners.Based on the research of automatic question and answer in the medical field,this paper uses 180 thousand medical question and answer data set to construct a question and answer retrieval library.The back-end uses Django framework,the web uses Vue,and the mobile end uses Java to implement the back-end server and front-end question and answer respectively.The webpage and Android medical Q&A APP have been compiled,and have been put into operation.