With the rapid improvement of people ’s living standards,there have been many problems,among which the most worthy of attention is health.For example,young people have problems such as malnutrition or obesity due to partial eclipse,picky eating and snacking.Middle-aged people have many sub-health problems due to excessive life pressure and irregular life.The elderly will have many health problems due to the decline of physical function.It is urgent to solve health problems.Early diagnosis and early treatment,efficient and accurate medical diagnosis will become an indispensable part of people ’s lives.In the above context,a large number of medical diagnostic systems emerge.However,there are some common problems in traditional medical diagnosis systems,such as the system requires a large number of doctors to maintain,the effect of medical diagnosis depends entirely on the individual level of the doctor,and so on.Therefore,this paper attempts to use natural language processing technology to enable medical diagnosis,and designs and implements an intelligent medical diagnosis auxiliary system based on BERT,aiming to promote the faster and better development of medical diagnosis system.The main work of this paper is as follows :1)A scheme of combining natural language processing with medical diagnosis to solve the common problems of traditional medical diagnosis system is proposed.At the same time,the training method of BERT + downstream task is used to train the diagnostic model.Medical diagnosis is ultimately a classification problem,that is,judging what kind of disease the patient is based on symptom information.Support vector machine and artificial neural network are two good classification algorithms.This paper uses different classification algorithms to design two diagnostic models for downstream tasks.2)An optimized MDBERT model is designed based on BERT native model and medical diagnosis scenario.The MDBERT model optimizes the BERT model from four aspects : First,a text corpus more suitable for medical diagnosis is selected for pre-training tasks.Secondly,a digital integrated training method is adopted for MLM pre-training tasks.Thirdly,the neural network parameters are L2 regularized,which reduces the degree of overfitting of the model to a certain extent.Fourthly,the training method of mean_max_pooled is adopted for downstream tasks.3)Based on the diagnosis model and the above optimization scheme,the intelligent medical diagnosis auxiliary system is completed.The system realizes the friendly interaction between patients and doctors,such as patients ’ medical records,doctors ’ pre-diagnosis medical records(this function uses a diagnostic model,which can assist doctors in diagnosing medical records),doctors ’ diagnosis medical records,etc.At the same time,the system has been tested in many aspects,such as functional test and non-functional test,and it is concluded that the system basically meets the expected effect. |