A Knowledge Graph Enhanced Topic Modeling Approach for Herb Recommendation

Traditional Chinese Medicine (TCM) plays an important role in Chinese society and is an increasingly popular therapy around the world. A data-driven herb recommendation method can help TCM doctors make scientific treatment prescriptions more precisely and intelligently in real clinical practice, which can lead the development of TCM diagnosis and treatment. Previous works only analyzing short-text medical case documents ignore rich information of symptoms and herbs as well as their relations. In this paper, we propose a novel model called Knowledge Graph Embedding Enhanced Topic Model (KGETM) for TCM herb recommendation. The modeling strategy we used takes into consideration not only co-occurrence information in TCM medical cases but also comprehensive semantic relatedness of symptoms and herbs in TCM knowledge graph. The knowledge graph embeddings are obtained by TransE, a popular representation learning method of knowledge graph, on our constructed TCM knowledge graph. Then the embeddings are integrated into the topic model by a mixture of Dirichlet multinomial component and latent vector component. In addition, we further propose HC-KGETM incorporating herb compatibility based on TCM theory to characterize the diagnosis and treatment process better. Experimental results on a TCM benchmark dataset demonstrate that the proposed method outperforms state-of-the-art approaches and the promise of TCM knowledge graph embedding on herb recommendation.