Data-Driven Traditional Chinese Medicine Clinical Herb Modeling and Herb Pair Recommendation

As an important branch of medical field, Traditional Chinese Medicine(TCM) continues to be explored in data mining research. Taking advantage of machine learning models and deep learning methods, researchers dive into symptom analysis, disease prediction and medicine law. The combination of TCM herbs is the essential basis for compatibility of clinical prescriptions and its research has attracted plenty of attention. However, literature on herb recommendation for clinical diagnosis, to our best knowledge, is slightly lacking. The clinical herbs collocation will be chosen by doctors in consideration of not only the characteristics and pharmacodynamics of the herbs, but also the mutual effects formed with other herbs. Based on the real clinical prescription data, this paper constructs an analytical model to represent the relationship between prescription herbs and syndromes, and develops herb recommendation model. Firstly, by constructing a modeling process based on the LDA topic model, this paper shows the analysis model and presentation method for prescription herbs. Then, based on the mentioned modeling, we propose a double-end fusion recommendation framework, including methods of adjusting weight proportion and similarity remapping. This research conducts experiments on relevant outpatient medical record data, which confirm that the proposed model can reflect the basic principles of herb combination in clinical diagnosis and the proposed fusion recommendation model has good performance in evaluation metrics.