<p>With the rapid advancement of modern herbal prescription recommendation (HPR) methods powered by artificial intelligence (AI), existing models for HPR often overlook the personalized attributes of patients, disregarding crucial factors in the clinical diagnosis and treatment processes of traditional Chinese medicine (TCM). To address this issue, we introduce a novel knowledge graph (KG) diffusion model for personalized TCM recommendation (TCM_DiffPR). Firstly, we utilize the patient’s personalized attribute information to derive symptom representations through ’prompt fine-tuning,’ further enhancing these prompts with a pre-trained model using contrastive learning (CL). Secondly, the algorithm integrates a generative model with a data augmentation paradigm grounded in KG diffusion, addressing the misinterpretation of symptom preferences within the KG to achieve robust knowledge acquisition. Finally, we introduce a collaborative KG convolution mechanism, combining collaborative signals that reflect symptom-herb interaction patterns to guide the KG diffusion process. Extensive experimental comparisons were conducted on two public datasets and a clinical dataset, revealing that our algorithm surpasses existing HPR methods and serves as an effective reference point for clinical diagnosis and decision support.</p>