Latent Semantic Analysis Of Clinical Data In Traditional Chinese Medicine

Traditional Chinese medicine(TCM)is an important part of Chinese traditional culture.It has been used to treat a variety of complex diseases and has achieved convincing results.Under the background of big data,a large number of medical data have been accumulated in the clinical practice of TCM.Modeling and analysis of these data can be used for clinical auxiliary diagnosis and treatment,and promote the theoretical and clinical development of TCM.Therefore,how to integrate TCM theory with clinical data modeling is a research emphasis.In addition,medical data contain complex semantic relationships of TCM entities,how to use these data to improve the effectiveness of the model is also a problem.Therefore,this paper proposes a latent semantic analysis technique for TCM clinical data.Specifically,this paper explores the relationship between symptoms and herbs in TCM medical cases data and recommends herbs for a given group of symptoms through improved topic modeling methods.First,a Multi-Content LDA model is proposed,which uses the concept of pathogenesis in TCM theory to analyze the relationship between symptoms and herbs in medical cases.And the corresponding herb recommendation method is proposed.Then,two embedding methods of TCM entities are proposed based on various forms of TCM data,which can be used to provide TCM entity embeddings with more semantic information.Furthermore,a Multi-Content embedding LDA model that integrates embeddings into topic model is proposed.This model is more effective on analysis and has better recommendation performance.Main contributions of this paper are summarized as follows:· TCM diagnosis and treatment based on Multi-Content LDA model A MultiContent LDA model(MC-LDA)is proposed,which considers the pathogenesis inTCM theory as latent topic in topic model.The model is used to connect the symptoms and herbs in the medical cases data.The output of the model can be used to analyze the relationship between symptoms and herbs.In addition,a herb recommendation algorithm is proposed for auxiliary diagnosis and treatment.Given a group of patient's symptoms,a recommended group of herbs can be obtained.· Embedding methods of TCM entities Two embedding methods of TCM entities are proposed,and the results are visualized and analyzed.The first method is based on context information,while the other is based on TCM knowledge graph.The embeddings of TCM entities are based on more abundant medical data and contains more abundant information.These embeddings can be used as the input of machine learning models to enhance the effect of the models.· TCM diagnosis and treatment based on Multi-Content embedding LDA model Considering the semantic information existing among TCM entities,a Multi-Content embedding LDA model(MC-eLDA)is further proposed.Each medical document contains a set of symptom embeddings and a corresponding set of herb embeddings,which are modeled by Gaussian distributions.In this way,the semantic relevance of the symptoms and herbs in each topic is improved,then the effect of the model mining the relationship between symptoms and herbs and the effect of herb recommendation are improved.