A Deep Learning-Based Herb Pair Discovering Approach

Abstract

The use of artificial intelligence methods to assist in the discovery of Traditional Chinese Medicine (TCM) pairs provides practical significance for the inheritance, development, and innovation of TCM. Most of the current herb pair mining methods only consider the attribute information of a single decoction piece and are based on the existing single machine learning model, so the quality of herb pair discovery is not high. This paper uses deep learning methods to solve the problem of TCM herb pair discovering. The properties of nature, taste, and meridian of the decoction composing the herb pair were added to the data set, and the knowledge-enhanced semantic representation (ERNIE) pre-training model was employed based on context and location information. We also compared it with the herb pair discovering effect of CNN, RNN, BERT, ERNIE models and common classification models. The results show that ERNIE can find potential herb pair effectively in the collection of TCM decoction pieces, and has a high effectiveness.