Indication Discovery Of Chinese Herbs Based On Information Recommendation

Chinese herb is the basic unit of Traditional Chinese Medicine(TCM)clinical prescription and the key that TCM doctors provide treatment plans for patients.The clinical characterization of TCM treatment(it s main connotation are symptoms and signs),that is,the personalized symptoms of patients are the important basis and target of clinical prescription medication,and also the important knowledge to support the personalized clinical treatment of TCM.Doctors rmodify prescription according to symptoms mainly dependent on common sense or empirical association between herb and symptom.However,because of the complexity of the efficacy and chemical composition of herbs,the discovery and research of the association between herb and symptom is still a complex problem.Therefore,on the basis of data accumulation in the current field and the progress of analytical methods,the research of appropriate methods for the analysis of herb-symptom association is of great value for personalized clinical diagnosis and treatment and the research and development of new Chinese patent medicine.Based on the literature data of Chinese pharmacopoeia and Chinese materia medica,this article carried out the research on the knowledge discovery of herb-symptom association based on collaborative filtering,matrix factorization and network representation learning.The main work includes the following two aspects:(1)This article researches the discovery methods of herb-symptom associations based on similarity calculation.Firstly the article constructs a standard data set based on the manual processing of Chinese pharmacopoeia and Chinese materia medica by artificial processing,which includes 48,000 herb-symptom associations,providing a data basis for the research of the relevant methods.On the basis,the article researches two methods of herb-symptom associations prediction based on similarity calculation:collaborative filtering and network embedding learning.The two methods fully considered the attribute’s information of herb such as taste,efficacy.Research shows that,among the methods based on collaborative filtering,the performance of feature fusion model is more significant(the coverage is improved from 44%to 47%),among the methods based on network embedding representation learning,the Deep Walk-based method has better performance.In general,the collaborative filtering method of feature fusion has the best performance among the knowledge discovery methods of herb-symptom associations.(2)This article conducts the research of the analysis of herb-symptom associations using matrix factorization.Firstly,the article proposes the methods based on non-negative matrix factorization(NMF)and singular value decomposition(SVD)to analyse herb-symptom associations.The result shows that the performance of SVD-based method is better than NMF-based method.In view of the sparsity of the existing herb-symptom associations,this article uses the functional similarity between herbs to supplement the matrix of herb-symptom association.Analysis result shows that matrix factorization methods with association completion have been improved to a certain extent(RMSE reduces from 0.8709 to 0.7122).In addition,for the further use of a variety of attribute information of herb,this article proposes a SVD method based constructing vectors using network embedding learning algorithm according the meta-path to analyse herb-symptom associations.This method has the best performance(the RMSE is reduced from 0.7122 to 0.3062 compared with the other methods).The above results suggest the important role of the relevance functional attributes of herb in the research of the analysis of herb-symptom association.