Construction and Application of Traditional Chinese Medicine Knowledge Graph Based on Large Language Model

Area of Science:

• Computational linguistics
• Artificial intelligence
• Traditional Chinese Medicine

Background:

• Traditional Chinese Medicine (TCM) possesses extensive knowledge but faces challenges in modernization and information extraction.
• Developing knowledge-based services for TCM requires structuring its complex and diverse information system.
• Integrating historical texts with open-source data is crucial for TCM advancement.

Purpose of the Study:

• To propose and develop a Large Language Model (LLM)-driven approach for structuring TCM knowledge.
• To enhance semantic understanding and knowledge extraction within TCM contexts using a fine-tuned LLM.
• To create an intelligent TCM Q&A system and a knowledge graph for improved diagnosis and treatment.

Main Methods:

• Developed a Fine-Tuning ChatGLM3-6B (FT-ChatGLM3) model optimized for Chinese-language processing on the AliCloud DSW platform.
• Integrated historical TCM texts with open-source TCM datasets for model training.
• Developed a BERT-based TCM Entity Recognition (TCMER) model and constructed a knowledge graph using FT-ChatGLM3 outputs.

Main Results:

• FT-ChatGLM3 demonstrated strong performance in TCM applications, providing accurate diagnosis and therapeutic suggestions.
• The intelligent TCM Q&A system significantly improved accuracy and efficiency in diagnosis and recommendations.
• The TCMER model effectively systematized and structured TCM knowledge, enhancing retrieval and consistency.

Conclusions:

• The LLM-driven approach, integrating FT-ChatGLM3 and TCMER, accelerates TCM knowledge graph development.
• This methodology advances the modernization and intelligent application of TCM in global healthcare.
• The study successfully structured TCM knowledge, improving information accessibility and application for healthcare professionals.