Research On Traditional Chinese Medicine Auxiliary Diagnosis And Treatment Methods Based On BERT

Traditional Chinese medicine(TCM)is my country’s traditional medicine and an important part of modern medicine.Diagnosis and treatment of patients by TCM physicians usually rely on long-term clinical experience and professional ability,which leads to problems such as strong subjectivity and uneven treatment levels in the process of diagnosis and treatment of patients by TCM physicians.Deep learning technology is widely used in different fields,in the medical field,the technology can assist doctors in clinical decision-making.Using the TCM symptom texts and TCM prescriptions collected in this research topic,this thesis studies the BERT model in deep learning,and applies it to the standardization of TCM symptoms,auxiliary diagnosis of TCM diseases and prescription recommendation,reducing the need for TCM physicians in the diagnosis and treatment process.The main work of this thesis is as follows:(1)For the TCM symptom normalization task,this thesis proposes a BERT-based TCM symptom normalization method.This thesis proposes a BERT-based TCM symptom normalization method.First,the semantic representation of symptom words is obtained using the BERT model;then,the similarity between two different symptoms is calculated by Euclidean distance;finally,the symptoms with higher similarity are normalized.Standardizing polysemy in TCM symptom texts can reduce the data redundancy problem in TCM symptoms,and is the basis for the research on TCM auxiliary diagnosis and prescription recommendation methods.(2)For the task of TCM clinical disease diagnosis,this thesis proposes an auxiliary diagnosis method of TCM diseases that can enhance the local feature extraction of symptom text.First,use BERT to obtain the vector representation of TCM symptom text;then,perform convolution pooling operation to obtain a global vector,and integrate the global vector into the extraction of local features;The fully connected layer and the softmax function complete the prediction of the patient’s disease.The experimental results show that the TCM disease auxiliary diagnosis method proposed in this paper has better performance than the comparison method.(3)For the task of TCM clinical prescription,this thesis by changing the mask mechanism of the attention matrix in BERT,a single BERT is realized to complete the task of TCM prescription generation.In the training phase,the symptom text and the TCM prescription are respectively input into the BERT model,and the attention matrix of the prescription sequence in the model is masked,so that BERT can learn the ability to generate TCM prescriptions.In the test phase,the patient’s symptom text is input,and the beam search algorithm is used to decode the prescription sequence to generate the TCM prescription corresponding to the patient’s symptoms.The experimental results show that the prescription recommendation method proposed in this paper can effectively provide reference for TCM physiciansThis thesis first standardizes the symptoms of TCM,and then models the diagnosis and prescription process in the process of TCM diagnosis and treatment,so as to realize the auxiliary diagnosis and treatment of TCM patients’ diseases,and verify the effectiveness of the proposed method through experiments.It can provide reference for TCM physicians in the process of clinical diagnosis and treatment,reduce mistakes made by TCM physicians in clinical practice,and improve the quality and efficiency of TCM clinical diagnosis and treatment.