Research Of The Symptom-Syndrome-Drug Relationship Based On Attention Mechanisms

Traditional Chinese medicine(TCM)is an important part of world medicine.The treatment based on syndrome differentiation is the basic principle of Chinese medicine clinical practice and the embodiment of the combination of theory and practice.Studying the principle of the syndrome differentiation treatment is to explore the internal connection of symptoms-syndrome-drugs of TCM.At present,due to the diversity and complexity of patients' clinical symptoms and the differences in doctors' diagnosis and treatment experience,it has brought serious challenges to the accurate grasp of clinical syndrome differentiation.At the same time,there are differences in the diagnosis and treatment ideas of various syndrome differentiation schools,resulting in different choices of treatment drugs,and it is difficult to grasp the rules of administration.Therefore,with the help of algorithms such as attention mechanism,deep learning technology and subspace clustering,this paper constructed different task models to explore the symptom-syndrome-drug correspondence relationship of traditional Chinese medicine,and formed an automated syndrome diagnosis algorithm to provide clinical syndrome differentiation treatment decision support.Moreover,we also explored the corresponding relationship between symptoms and drugs at the global pathological level,excavates the internal rules,determines the administration and administration specifications,and enhances the clinical diagnosis and treatment ability of traditional Chinese medicine.The main research contents of the article are as follows:(1)We proposed the syndrome classification model based on attention mechanism.In view of the diverse symptoms of patients,it is difficult to determine the symptoms of the disease.Based on the neural network theory,this paper proposed a syndrome diagnosis algorithm combining attention mechanism and multi-layer perceptron.The attention mechanism intelligently discriminated the correlation between symptoms and syndromes and contributed weight to the reasonable distribution of symptoms.The multi-layer perceptron could accurately learn the corresponding relationship between key symptoms and syndromes according to the weighted symptom vectors,and quickly classified syndromes.Experimental results showed that our model not only improved the accuracy of syndrome diagnosis,but also selected the representative core symptom group for syndromes.(2)We proposed the syndrome classification model based on multi-instance and multi-view learning.Aiming at the problem of insufficient utilization of patient's symptom information,it is difficult to improve the effect of syndrome classification.Based on the attention mechanism,this paper proposed a classification model combining multi-instance and multi-view learning.The multi-instance generator generated a variety of potential key symptom combination instances;the multi-view learning module used the convolution kernel to learn the global representation of symptoms,and created an interactive representation spectrum function to learn adaptive correlation representation from the interactive information between symptoms;Attention mechanism fused instance information,and extracted high-level integrated information for syndrome diagnosis.Experiments on the dataset showed that the classification accuracy of the model reached 80.96%,which surpassed most advanced algorithms and proved its effectiveness.(3)We proposed the case information integration and clustering system based on recurrent neural network.In order to explore the relationship between the symptoms and drugs of different patient groups of the same disease,this paper built a multi-model representation framework and combines subspace clustering algorithms to cluster patients.The framework used one-hot coding,word embedding model and Bi-LSTM model to provide high-level semantic information from the patient's basic attributes and diagnosis and treatment sequence data to form a patient's body condition representation vector.In this paper,the sub-space clustering algorithm based on spectral clustering was used to cluster the patient's representations,and the population clusters of different body conditions are divided to explore the correspondence between the symptoms and drugs of different populations' Drug pairs and drug groups,to summarize the symptoms of different populations and the hidden rules of drug connection,to provide auxiliary reference for clinical specific diagnosis and treatment.