通讯作者: 叶均明, 47480355@qq.com

DOI：10.12201/bmr.202608.00020

Design and Implementation of a Hot-swappable Medical Knowledge Base System Based on Hybrid Retrieval Strategy

• 摘要：目的/意义 构建支持热插拔的多格式医学文档知识库系统，实现从精准知识检索到专业知识问答的一体化服务，以提升知识检索效率与应用便捷性。方法/过程 于本地构建支持多格式异构文档知识库，基于动态权重混合检索策略，完成知识库文档的标准化的解析提取及个性化的即插即用；利用检索增强技术（RAG）融合本地医学知识库与大语言模型，通过多级缓存、流式响应等机制，实现高效、可溯的医学专业问答生成系统。结果/结论 在自建的多格式医学知识测试集进行验证，结果表明，相较于单一的向量检索或关键词检索策略，本系统检索精准率显著提升，并借助大语言模型生成能力，可高效满足医务人员的即时知识获取与智能问答的迫切需求。

  关键词： 医学知识库；热插拔；混合检索；检索增强生成；智能问答; ; ; ; 

   

  Abstract: Objective: To establish a medical record generation intelligent body with a large language model as the core to realize the intelligent generation of electronic medical record instruments and reduce the pressure of clinical medical record writing. Methods: Through data cleaning and locally deployed large language model simulation for dialogue generation, the data reserve for model fine-tuning is completed, and model fine-tuning technology is utilized to train the general large language model into a vertical model for medical record generation, meanwhile, a local knowledge base is established based on standards such as medical record writing specifications, and finally, the medical record generating agent is built with the fine-tuned model. Results: The evaluation results by model fine-tuning show that the medical record instruments generated by the large language model after supervised fine-tuning are more accurate and more in line with the requirements of medical record instrument specifications. By designing the workflow of intelligibles with retrieval of local knowledge base retrieval correction, it can make the generated medical record text more in line with the requirements of medical quality control. Conclusion: This study combines multiple technologies to construct a medical record generating agent, and the generated electronic medical record documents have higher professionalism and reliability, which provides empirical reference and empirical support for the application and development of intelligent medical record generation.

  Key words: medical knowledge base; hot swapping; hybrid retrieval; retrieval-augmented generation; Intelligent Question and Answer

  提交时间：2026-08-07

  版权声明：作者本人独立拥有该论文的版权，预印本系统仅拥有论文的永久保存权利。任何人未经允许不得重复使用。
• html

• 图表

• 车纯凯, 李亚子. 基于混合检索增强生成Hybrid RAG的卫生统计年鉴问答系统的研究与构建. 2026. doi: 10.12201/bmr.202604.00082

  何剑虎, 汤琳琳, 伊胜月, 杨春波, 王德健. 基于检索增强生成的妇产科医生助手智能体构建与应用研究. 2026. doi: 10.12201/bmr.202606.00073

  惠婷, 申艳妮. 面向语义的电子病历智能文本检索技术的研究与设计. 2025. doi: 10.12201/bmr.202512.00082

  吕婷钰, 李晓瑛, 刘宇炀, 杜晋华, 李心怡, 罗妍, 唐小利, 任慧玲, 刘辉, 尹浩. 中文医学知识大模型问答语料数据集构建研究. 2024. doi: 10.12201/bmr.202404.00002

  徐远虹, 赖书兰, 张与桐, 关昌昊, 胡红濮. 面向基层医疗的糖尿病知识库设计与规则驱动应用框架研究. 2026. doi: 10.12201/bmr.202608.00017

  龚宇新, 李俊豪, 向菲, 张兰. “AIGC+医学文献检索”课程优化设计与实践路径研究. 2025. doi: 10.12201/bmr.202504.00037

  杨亚洲, 郑石林, 周小毅, 章施静, 赵坚. 基于知识图谱的HIV防治大模型RAG问答系统构建研究. 2025. doi: 10.12201/bmr.202511.00061

  吴欢, 何昆仑. 基于循证医学和电子病历数据的通用医学知识图谱构建. 2024. doi: 10.12201/bmr.202409.00027

  孙月萍, 王娟, 董良广, 刘燕, 杨丽, 李姣, 侯丽. 基于预训练模型的医学习题解析半自动生成方法研究. 2026. doi: 10.12201/bmr.202602.00048

  申艳妮, 惠婷. 基于大语言模型的医疗机构图像智能化归档系统设计. 2026. doi: 10.12201/bmr.202602.00047

• 序号	提交日期	编号	操作
  1	2026-01-22	10.12201/bmr.202608.00020V1	下载

• 公开评论  匿名评论  仅发给作者 提交评论