Scholarlink

With the advent of large language models (LLMs), evaluating their diagnostic thinking abilities is gaining much consideration in the field of traditional Chinese medicine (TCM). Within this context, we propose an innovative framework named CotKE-TCM, which integrates chain-of-thought and knowledge retrieval enhancement methodologies for TCM diagnostic thinking. This framework integrates three essential medical knowledge resources: 1) acquiring case knowledge through case similarity analysis to offer intuitive reference cases for TCM diagnosis; 2) leveraging LLMs to produce specialized TCM knowledge, guiding the model towards a more accurate understanding of disease causality and syndrome differentiation; 3) obtaining clinical experience knowledge via retrieval techniques to effectively complement the limitations inherent in training data. By skillfully combining these three knowledge sources, we have developed a comprehensive prompting engineering system. The experimental outcomes demonstrate exceptional performance on the final test dataset provided by the 10th China Health Informatics Processing Conference (CHIP 2024). This not only verifies the efficacy of the CotKE-TCM framework but also provides significant references and insights for the future advancement of intelligent TCM.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A TCM Syndrome Differentiation Thinking Method Based on Chain of Thought and Knowledge Retrieval Augmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jianfeng Zhang,
  • Xiang Li,
  • Jian Fang,
  • Wenqi Wei,
  • Menglin Cui

摘要

With the advent of large language models (LLMs), evaluating their diagnostic thinking abilities is gaining much consideration in the field of traditional Chinese medicine (TCM). Within this context, we propose an innovative framework named CotKE-TCM, which integrates chain-of-thought and knowledge retrieval enhancement methodologies for TCM diagnostic thinking. This framework integrates three essential medical knowledge resources: 1) acquiring case knowledge through case similarity analysis to offer intuitive reference cases for TCM diagnosis; 2) leveraging LLMs to produce specialized TCM knowledge, guiding the model towards a more accurate understanding of disease causality and syndrome differentiation; 3) obtaining clinical experience knowledge via retrieval techniques to effectively complement the limitations inherent in training data. By skillfully combining these three knowledge sources, we have developed a comprehensive prompting engineering system. The experimental outcomes demonstrate exceptional performance on the final test dataset provided by the 10th China Health Informatics Processing Conference (CHIP 2024). This not only verifies the efficacy of the CotKE-TCM framework but also provides significant references and insights for the future advancement of intelligent TCM.