Traditional Chinese Medicine Prescription Recommendation Technology Using Syndrome Information

Traditional Chinese Medicine(TCM)is a unique traditional medicine that focuses on nursing constitution and prevention,which has been passed down from generation to generation.In recent years,artificial intelligence technology represented by deep learning has developed rapidly,which injecting new vitality into the development of TCM towards intelligence and informatization.Developing a recommendation model for TCM based on deep learning technology,which can achieve intelligent diagnosis based on symptoms and dispensing prescriptions,is one of the current research hotspots.Dialectical treatment is the basic principle for treating diseases in TCM,that is,considering the symptoms collected through "inspection,listening and smelling,inquiry,palpation and pulse taking" as a whole,diagnosing the cause and inducing it into a certain syndrome.Then determining a treatment plan based on the syndrome,and prescribing medicine.As a bridge between diseases and herbs,syndromes play a crucial role in the process of prescribing medicine.Capturing the relationship between symptoms,syndromes,and herbs can help make more reasonable recommendations for TCM.However,due to the ambiguity and complexity of dialectics itself,most real prescriptions have the phenomenon of missing labeled with syndrome.TCM emphasizes the compatibility of herb pairs.The high-frequency herb combinations for a certain syndrome often reflect the close relationship between them,which can help explain and analyze the syndrome.In summary,the task of TCM recommendation incorporating syndrome information proposed in this article mainly faces the following challenges:(1)We need to design a method for explicitly modeling the relationship between symptoms,syndromes and herbs.(2)We need to mine the relationship between syndromes and commonly used herb combinations while predicting syndromes based on prescription information.The main work of this article includes the following aspects:1.Syndrome-aware herb recommendation with heterogeneous graph neural network.To model the relationship between symptoms,syndromes and herbs explicitly,we propose a syndrome-aware herb recommendation with heterogeneous graph neural network.Firstly,clustering is used to alleviate the negative impact of uneven distribution of syndrome data on downstream models.Subsequently,multiple disease-herb subgraphs are constructed based on cluster numbers,and syndrome information are integrated into the graph convolution message propagation process of the subgraph to learn the representation of herb,symptoms,and syndrome classes.Finally,we construct a syndrome-herb graph to model the relationship between syndromes and herbs.By considering the node representation of two semantic spaces comprehensively,the multiple interactions between syndromes and herbs under different syndromes were captured to recommend herbs.The experimental results on a real Chinese medicine dataset demonstrate the effectiveness of this model,which is helpful for grasping the herb rules of diseases under different syndromes.2.Explainable syndrome prediction based on binary herb combinations.Aiming at mining the relationship between syndromes and commonly used herb combinations,we propose an explainable syndrome prediction model based on binary herb combinations.Firstly,from the perspective of binary herb combinations,the association between binary drug combinations and syndromes is defined.Then,initial values are generated based on the idea of TF-IDF,and the multilayer perceptron is used to optimize the parameters.Finally,the binary herb combinations in the prescription are analyzed,and the association values are summarized for syndrome prediction.The experimental results validate the accuracy of the model and can highlight binary herb combinations with high scores as a reference for predicted syndromes.3.Herb recommendation and analysis system integrated with syndrome information.We integrate the above two models,fuse prescription information and TCM science popularization information,and build a herb recommendation analysis system incorporating syndrome information.The system provides functions such as herb recommendation,syndrome prediction and analysis,and expands the analysis of similarities and differences in medication under multiple syndromes.It shows the contribution of binary herb combinations to predicted syndromes with the form of heat maps.This system may provide users and TCM researchers with more convenient services.