Personalized Herb Recommendation in Traditional Chinese Medicine

Introduction: Traditional Chinese Medicine (TCM) continues to gain global recognition for its personalized treatment options for chronic diseases. Accurate TCM recommendations are the foundation of effective prescription formulation; however, existing computational models still struggle to achieve this goal. These existing systems suffer from fundamental limitations, such as sparse coding of symptoms, inadequate semantic representation, and a lack of context modeling. Methods: To address these bottlenecks, we introduce HMHRec, a deep learning framework driven by multiple attention mechanisms and highway networks. First, patient symptom sets and herb sets are encoded as multi-hot vectors, which are then transformed into dense embeddings via a trainable matrix. The highway network controls information flow to enhance nonlinear feature representation, while the attention module parses complex co-occurrence patterns and hidden relationships between symptoms. Finally, the Top-K selection strategy sifts through the processed data to extract the most appropriate herbs. Results: We use two publicly available datasets, TCM1 and TCM2 to comprehensively evaluate the effectiveness of the framework and compare it with representative baseline models. Experimental results show that HMHRec consistently outperforms the baseline models. On the TCM1 dataset, the proposed model achieves a relative improvement of 7.53%, 7.65%, and 5.16% on the indicators F1-Score@5, F1-Score@10, and F1-Score@20, respectively. Similar improvements were observed on the TCM2 dataset, achieving 7.04%, 4.37%, and 4.12% improvements on the same metrics, respectively. Discussion: By seamlessly integrating the highway network with multiple attention mechanisms, the HMHRec model successfully bridges the semantic gap between a specific symptom and the corresponding herb. Both experiments and case studies highlight the practical value of the framework, which proves that it can provide data-driven and highly personalized herbal medicine recommendations for clinicians in real TCM practice. Conclusions: The HMHRec model can generate highly accurate and clinically meaningful herbal medicine recommendations, which effectively overcome the limitations of previous models. This advanced framework has great potential to become a powerful clinical decision support tool, which can help clinicians provide evidence-based and personalized TCM treatment plans.