一种面向中医问诊的数据知识双驱动大模型OA

Generalized large language models(LLMs)in the field of Chinese medicine consultation suffer from the problems of insufficient knowledge retrieval ability and low query analysis accuracy,which affect their accuracy and reliability.Therefore,this study proposes a data-knowledge dual-driven large language model for TCM consultation(DualKAG-ShenNong).DualKAG-ShenNong adopts a medical-style fine-tuning strategy,which introduces an external knowledge base during fine-tuning,to make up for the limitations of data-driven models in terms of knowledge utilization.At the same time,it combines the knowledge graph(KG)-based retrieval augmented generation(RAG)technology to provide inference chains with TCM theoretical knowledge and design prompt templates,which fully utilizes the powerful genera-tive and inference capabilities of the LLM to provide users with accurate TCM consultation services.A large number of experimental results show that DualKAG-ShenNong outperforms the traditional model in a number of metrics such as ROUGE,BLEU,etc.,which verifies its effectiveness and scientificity in Chinese medicine intelligent Q&A,and provides a solid application support.