Traditional Chinese Medicine Assisted Inquiry System Based On Task-based Dialogue Research And Implementation

TCM diagnosis is the foundation of TCM diagnosis and treatment,and it is also the most critical part of the TCM diagnosis and treatment process.The main methods of TCM diagnosis are inspection,auscultation,interrogation,and palpation.Among them,interrogation is an important part of TCM diagnosis methods.Through multiple rounds of dialogues between TCM doctors and patients,interrogation can help understand the occurrence of patients’ diseases,development,existing symptoms,etc.In the process of TCM consultation,the content and order of consultation are often determined by the experience of TCM physicians.The current combination of task-based dialogue and TCM consultation is still in the development stage,and no personalized TCM consultation model has been established.Therefore,this thesis takes the task-based dialogue of TCM consultation as the research object,carries out research on natural language understanding of TCM consultation,natural language generation of consultation,and research on TCM syndrome identification,and designs and implements a TCM consultation platform based on task-based dialogue.The main work is as follows :1.For the existing natural language understanding models,there is no effective combination of slot filling tasks and intent recognition tasks,and the existing models have information forgetting problems in long text dialogues.A natural language understanding joint model(Joint Model of Intention recognition and Slot filling based on BERT+Bi-GRU+ Attention mechanism,JMIS-BBA)based on the BERT+Bi-GRU+Attention mechanism is proposed,training the intent recognition and slot filling in current natural language understanding task together,introducing Bi-GRU+ attention mechanism to improve the recognition accuracy of long text.The experimental results show that the Slot-F1 score,Intent-F1 score,and Sentence accuracy of JMIS-BBA in the SNIPS(Spoken Natural language processing in service of Intelligent Personal Assistants,SNIPS)dataset and ATIS(Airline Travel Information System,ATIS)dataset are 97.8%,97.6%,92.8% and 98.0%,97.6%,92.0%,respectively.The Slot-F1 score,Intent-F1 score,and Sentence accuracy in the TCM clinical data set are 96.3%,100.0%,and 100.0%,which provides an effective method for TCM Natural language understanding of medical inquiries.2.For the existing model,the fixed encoding method is adopted to encode the dialogue actions into one-hot form.Although good results have been achieved in the fixed dialogue actions,the new dialogue actions need to be re-encoded,and the model maintainability is poor;in addition,the existing models treat multiple slots equally,ignoring the influence of different slots on dialogue generation,and an Action and Slot pairs based Auto-Encoder conversation generation model(ASAE)is proposed,using the automatic encoder of dialogue action and slot pair to learn the behavioral features of the dialogue,and the decoder generates natural language responses according to slot pairs with different weights.The test results show that the BLEU-4 scores of ASAE in the Hotel,Restaurant,and Laptop fields of the SF(San Francisco,SF)dataset are 68.9%,80.1%,and 42.3%,and the ERR scores are 2.6%,2.8%,and 2.7%,respectively.3.In view of the fact that existing models ignore temporal and spatial feature information and unsupervised information,a syndrome identification(Automatic Encoder and Convolutional Neural Network with unsupervised information,AECNN)that fuses unsupervised information is proposed.Using unsupervised information and supervised information,LSTM encoder and CNN network are introduced to capture the sequence and spatial features of TCM features,and learn the characteristics of TCM expert consultation experience.The test results show that AECNN can improve the accuracy of TCM syndrome identification,and the recognition accuracy rates on the CIFAR10 public dataset and the TDCC TCM clinical dataset are 74.1% and 95.3%,respectively.4.Through frameworks such as SpringBoot,MyBatis and Vue,based on BS(Browser/Service,B/S)architecture,using Java language,Python language,and My SQL database to design and implement a TCM consultation platform based on task-based dialogue,with intent recognition,slots filling,syndrome identification,and scene interaction have been simulated to achieve the effect of TCM consultation through functional testing and performance testing.