Zhang Bo, Li Ruifang, Yin Kedong, Hua Shuo, Li Shiyu, Jiang Mengwan, An Haoping, Li Peng
Key Laboratory of Functional Molecules for Biomedical Research, Henan University of Technology, Zhengzhou, 450001, P. R. China.
College of Information Science and Engineering, Henan University of Technology, Zhengzhou, 450001, P. R. China.
Interdiscip Sci. 2025 Jul 2. doi: 10.1007/s12539-025-00735-1.
Traditional Chinese Medicine (TCM) is a vital component of the Chinese heritage, embodying a wealth of medical knowledge and distinctive therapeutic practices. A critical challenge in TCM modernization lies in extracting essential information from its complex and diverse knowledge system to develop knowledge-based services, which represents a cutting-edge research focus. This study proposes a Large Language Model (LLM)-driven approach for structuring TCM knowledge integrating historical TCM texts with open-source TCM datasets. A Fine-Tuning ChatGLM3-6B (FT-ChatGLM3) model was developed on the AliCloud DSW platform, optimized specifically for Chinese-language processing to enhance semantic understanding and knowledge extraction within TCM contexts. FT-ChatGLM3 powers an intelligent TCM Q&A system, significantly improving the accuracy and efficiency of diagnosis and therapeutic recommendations. Furthermore, a BERT-based TCM Entity Recognition (TCMER) model was developed, and a knowledge graph was constructed using FT-ChatGLM3's outputs. Experimental results demonstrate that FT-ChatGLM3 achieves strong performance in TCM applications, delivering precise diagnosis and treatment suggestions. The TCMER model also exhibits high efficacy, facilitating the systematization and structuring of TCM knowledge, while improving knowledge retrieval and consistency. The integration of FT-ChatGLM3 and TCMER not only accelerates the development of TCM knowledge graphs but also advances TCM modernization and its intelligent application in global healthcare.
中医是中华传统文化的重要组成部分，蕴含着丰富的医学知识和独特的治疗方法。中医现代化面临的一个关键挑战是从其复杂多样的知识体系中提取关键信息，以开发基于知识的服务，这是一个前沿的研究重点。本研究提出了一种由大语言模型驱动的方法，用于构建将中医历史文本与开源中医数据集相结合的中医知识体系。在阿里云DSW平台上开发了一个微调ChatGLM3-6B（FT-ChatGLM3）模型，该模型针对中文处理进行了优化，以增强中医语境中的语义理解和知识提取能力。FT-ChatGLM3为一个智能中医问答系统提供支持，显著提高了诊断和治疗建议的准确性和效率。此外，还开发了一个基于BERT的中医实体识别（TCMER）模型，并使用FT-ChatGLM3的输出构建了一个知识图谱。实验结果表明，FT-ChatGLM3在中医应用中表现出色，能够提供精确的诊断和治疗建议。TCMER模型也显示出很高的效能，有助于中医知识的系统化和结构化，同时提高知识检索能力和一致性。FT-ChatGLM3与TCMER的整合不仅加速了中医知识图谱的发展，也推动了中医现代化及其在全球医疗保健中的智能应用。
