TCM Entity Representation Leanrning And Herb Recommendation

Traditional Chinese Medicine(TCM)is a traditional medicine originated in China,which plays an important role in healthcare.Highly developed artificial intelligence,rep-resented by machine learning,is the key element in the promotion of TCM development A data-driven herb recommendation model based on machine learning can help TCM doc-tors make scientific treatment prescriptions more precisely and intelligently in real clinical practice,which can promote the development of TCM diagnosis and treatment.Previous works based on the topic model can not model the relationship between symptoms and herbs precisely only by co-occurrence.Therefore,introducing TCM domain knowledge into the herb recommendation model is better for modeling the relationship between symp-toms and herbs,and improves the performance of herb recommendationThis paper deeply studies how to introduce TCM domain knowledge in the herb rec-ommendation model based on topic model.First,we organize TCM domain knowledge by knowledge graph,so that we can use the knowledge efficiently.Then,we propose a representation learning method based on the heterogeneous network to obtain more ef-fective representations of TCM entities and support the herb recommendation model.At last,we propose a herb recommendation model based on topic model incorporating TCM entity embeddings and herb compatibility knowledge.The main work of this paper is summarized as follows·TCM knowledge graph We construct a TCM knowledge graph based on multi-sources heterogeneous data by using the top-down method.The knowledge graph can be used by downstream tasks such as herb recommendation·TCM Entity Embedding Model Based on Heterogeneous Network(TCME)Aim-ing at the problems of modeling heterogeneous information in TCM and high-level connectivity between TCM entities,we proposed TCME model to obtain the ad-vanced TCM entity embeddings for the herb recommendation task;TCM Entity Embedding Enhanced Topic Moldel based Herb Recommenda-tion(TCMETM)We propose TCMETM model by introducing TCM entity em-beddings into topic model for herb for effective modeling of the relationship be-tween symptoms and herbs.Furthermore,we propose HC-TCMETM model by in-troducing herb compatibility knowledge,make the result of herb recommendation more reasonable.