Construction of a Traditional Chinese Medicine Knowledge Graph Based on Multi-Agent Systems

With the rapid advancement of artificial intelligence technology, the digitization of traditional Chinese medicine knowledge systems remains an unresolved challenge, hindered by governance bottlenecks stemming from data complexity and fragmented knowledge. Traditional manual curation struggles to balance scale and precision, creating an urgent need for automated, high-accuracy knowledge graph construction solutions. This paper proposes a multi-agent-based method for constructing a knowledge graph of traditional Chinese medicine. By designing the multi-agent framework KGMAL, the knowledge graph construction task is decomposed into sub-tasks: entity extraction for each category and relationship construction. Each sub-task is handled by a specialized agent. The data used in this study is sourced from authoritative and comprehensive Chinese medicine encyclopedias, with detailed information on 10,268 types of Chinese medicinal materials collected via Python web scraping technology. Experimental results demonstrate that the construction of a traditional Chinese medicine knowledge graph based on the KGMAL framework achieves high accuracy. Comparative analysis reveals that triples constructed using DeepSeek-V3 as the base model exhibit the highest correctness rate, reaching 99%. Ultimately, a traditional Chinese medicine knowledge graph comprising 55,089 entities and 192,688 relationships was successfully constructed. The study confirms that the KGMAL framework significantly enhances the efficiency and accuracy of TCM knowledge graph construction. It provides robust technical support for addressing TCM data governance challenges, advancing the systematic integration of TCM knowledge, and facilitating its modern applications. This holds significant practical implications for promoting the innovative development of the TCM industry.