Research on Key Technologies of AI Large language Models in the Field of Traditional Chinese Medicine

With the accelerated development of traditional Chinese medicine (TCM) towards standardization, intelligence, and modernization, Large Language Models (LLMs) are providing key technical support for the diagnosis, treatment, and inheritance of TCM, owing to their inherent alignment with the holistic concept of "four diagnosis intergation". TCM-specific large language models (TCM-LLMs), as the core engine driving the intelligentization of TCM, have shown broad prospects in TCM knowledge organization, auxiliary diagnosis, and clinical decision support. First, this paper systematically reviewed the research progress of TCM-LLMs and proposed their construction process into four stages: corpus collection and preprocessing, construction of TCM dialogue datasets, model training and fine-tuning, and model evaluation. Although general-purpose foundation models (such as ChatGLM, Baichuan, and Ziya-LLaMA) possess strong language comprehension capabilities, they still face bottlenecks in the field of TCM, including poor knowledge timeliness, weak reasoning ability, and insufficient coverage of domain-specific expertise. Second, to address these challenges, TCM-LLMs must combine advanced techniques to enhance the domain-specific capabilities, including RAG, Knowledge Distillation, Prompt Engineering, SFT and RLHF. This paper systematically analyzes and elucidates the critical roles of these key technologies in TCM-LLMs. Finally, this paper presented summary and outlook on the future development opportunities and the many profound challenges facing TCM-LLMs.