Syndrome Differentiation In Traditional Chinese Medicine Based On Multi-Label Learning

With the rapid development of medical information technology,the development of intelligent medicine in the field of traditional Chinese medicine(TCM)is included in the projects encouraged by the state.Syndrome differentiation and treatment is the basic rule of TCM to identify and treat diseases.The establishment of a scientific and standardized intelligent method of syndrome differentiation is of great significance to the development of intelligent medicine in TCM.At present,the difficulty of intelligent TCM syndrome differentiation lies in the fact that there are multiple syndromes mixed in TCM clinical syndrome differentiation,which makes TCM syndrome differentiation essentially a multi-label learning task.At the same time,it is difficult to form a large standard dataset in the field of TCM due to the lack of TCM data collection standards and data privacy protection.It makes the size of TCM medical text data set small and the information distribution density sparse.In view of the above problems,the thesis thoroughly investigates the data mining and syndrome differentiation methods based on TCM medical data.The main work is as follows:(1)This thesis proposes a TCM syndrome differentiation analysis model based on improved ML-Relief F and multi-label deep forest.The sparse feature matrix generated by the medical text representation is filtered by an improved multi-label Relief F feature selection algorithm to obtain the optimal feature subset.The multi-label deep forest algorithm is introduced into the task of TCM syndrome differentiation analysis for the first time.By using its powerful representation learning ability,the syndrome prediction results are obtained by training small-scale medical texts.The experimental results prove the necessity of using improved ML-Relief F feature selection algorithm and the feasibility of the proposed model.(2)This thesis proposes a set of multi-source heterogeneous TCM domain knowledge graph(KG)construction processes based on Neo4 j.Due to the data accuracy defect of general medical KG,its clinical application is limited.Based on the medical record data of TCM prevention and treatment of chronic renal failure(CRF)and combined with external multi-source heterogeneous medical data,this thesis constructs KG of single disease based on TCM prevention and treatment of CRF from top to bottom.In the pattern layer,Protégé is used to construct the ontology structure of KG,and in the data layer,multi-source heterogeneous data is processed and fused to generate triples stored in the Neo4 j graph database.Through the visualization display and analysis of medical information,it is proved that the KG can be used in the medical scene with high requirement of knowledge accuracy.(3)This thesis proposes a multi-label TCM syndrome differentiation analysis model based on knowledge graph embedding.In order to better represent TCM medical texts,the embedded vector of external knowledge graph is used to provide additional hidden information.The representation learning process is divided into knowledge graph embedding module and text representation module.In order to integrate the two features reasonably and effectively,the KT-Fusion feature fusion method is proposed in this thesis.Finally,the multi-label deep forest algorithm is used as a classifier to obtain the prediction results of syndrome.Through comparative experiments,it is proved that the knowledge graph embedding model selected in this paper has the best embedding performance in the used data set,and the proposed model has obvious improvement in the multi-label classification performance of TCM syndrome differentiation analysis task.