Mongolian medicine prescription recommendation using graph attention networks leveraging semantic associations for precise predictions

The objective of this study is to address the challenges faced by traditional Mongolian medicine in the modern era. The complex knowledge system and challenges related to inheritance in Mongolian medicine represent significant obstacles to the modern development of this discipline. The present study introduces a graph attention network (GAT) model to address these issues. The GAT model establishes graphs of symptoms, Mongolian medicine, and symptoms-Mongolian medicine. The GAT model employs graph convolution operations to effectively capture the intricate relationships among symptoms and Mongolian medicines. This facilitates the model capacity to discern representations that are both discriminative of symptoms and Mongolian medicines. Consequently, the model is capable of matching appropriate Mongolian medicinal prescriptions according to the input symptoms. A series of experimental evaluations were conducted on a dataset derived from the Encyclopedia of Mongolian Medicine. These evaluations demonstrated that the proposed GAT model outperforms existing models in terms of prescription recommendation accuracy. Specifically, the model achieves an accuracy of 37.59%, representing significant improvements compared to other models. These findings suggest that the GAT model can effectively leverage the relationships among symptoms and Mongolian medicines to provide reliable prescription recommendations, offering a promising solution for the modernization of Mongolian medicine.