Chinese herbal medicine recommendations are a core part of personalized traditional Chinese medicine (TCM) diagnosis and treatment. However, the complexity of the multidimensional relationships in syndrome differentiation and treatment, herbal compatibility, and dosage selection poses significant challenges to clinical decision-making. Although artificial intelligence technology has made remarkable progress in TCM auxiliary diagnosis, a systematic review of Chinese herbal medicine recommendation methods remains lacking. This review aims to address this gap by systematically reviewing Chinese herbal medicine generation methods grounded in knowledge graph-based recommendations, deep learning-based recommendations, and hybrid model-based recommendations from 2016 to 2025. Major TCM databases that serve as foundational data sources, including traditional Chinese medicine systems, pharmacology database, and analysis platform, symptom mapping database, High-throughput Experimental and Reference Database, and Traditional Chinese Medicine Information Database, which are crucial for training these recommendation models. It further analyses their evolutionary technical patterns and clinical applicability, providing critical references for developing theoretically robust and clinically interpretable artificial intelligence models for TCM practice. Existing research focuses on constructing knowledge graph-driven Chinese herbal medicine recommendation models, which enhance the interpretability of recommendations by structuring the relationships among symptoms, Chinese herbal medicines, and diseases. Meanwhile, a clinical data-driven framework is introduced to discover potential patterns from real-world diagnosis and treatment scenarios. Deep learning-driven methods are adopted to achieve end-to-end feature learning for TCM knowledge reasoning. To improve the clinical applicability of Chinese herbal medicine recommendation models, a few studies have reported evaluation methods by experienced clinical doctors using herbal effectiveness and herbal compatibility scores to assess the reliability of the models and the accuracy of the recommendation results. Forming a comprehensive evaluation system may be the development trend of the evaluation system for clinical decision-support systems. This review outlines a theory-data-clinical ternary evaluation framework for Chinese herbal medicine recommendation models, providing a methodological innovation for developing intelligent systems that meet the standards of evidence-based medicine.
Yang et al. (Fri,) studied this question.