中文医学知识大模型问答语料数据集构建研究
Research on the Construction of a Question-Answer Corpus Dataset for Chinese Medical Knowledge Large Language Models
目的/意义 鉴于医学知识的专业性和复杂性,大语言模型在医学领域的问答任务效果不佳,所以需要定量评测大模型的回答效果以提高问答结果的准确性。构建中文医学知识问答语料数据集能够为医学垂域大模型提供标准化的评测基准,并以此改进大模型在处理中文医学问答任务中的准确率和效率。方法/过程 本研究构建了中文医学论文知识问答数据集、医学名词解释问答数据集和以中国执业医师考试真题(补充)为基础的问答数据集,并整理了相关开源数据集,用于评估大模型的医学知识覆盖范围、理解和生成能力。结果/结论 自主构建的中文医学知识问答语料数据集,丰富了现有中文医学问答语料的来源,能够作为一项标准化的评测基准,推动医学领域大模型实现客观全面的定量评估,今后将利用电子病历、在线健康社区等数据进行扩充,为健康中国战略的实施提供更坚实的AI支持。
Purpose/Significance Given the specialized and complex nature of medical knowledge, the performance of large language models (LLMs) in medical question-answering tasks is unsatisfactory. Therefore, its essential to conduct quantitative assessments of these LLMs performance to enhance the accuracy of LLMs in responding to medical domain questions. This paper focuses on the construction of a Chinese medical knowledge corpus dataset to enhance the accuracy and efficiency of LLMs in handling Chinese medical questions, which aims to establish a standardized evaluation benchmark for LLMs in the medical domain. Method/Process This study developed Q&A datasets encompassing Chinese medical paper knowledge, medical terminology explanations, and supplementary questions acquired from the Chinese medical licensing examination, as well as open-source Chinese medical Q&A datasets. These datasets would be useful for evaluating LLMs medical knowledge coverage, comprehension, and generation capabilities. Result/Conclusion The Chinese medical Q&A corpus datasets enrich the sources of existing data sets and promote the objective and comprehensive quantitative evaluation of large models in the medical field. In the near future, additional data such as electronic medical records and those from online health communities will be utilized to expand this dataset. All the efforts will offer stronger AI support for the Healthy China strategy.
李晓瑛、刘辉、唐小利、尹浩、任慧玲、罗妍、吕婷钰、刘宇炀、李心怡、杜晋华
医学研究方法医药卫生理论
大语言模型 语料数据集 模型评测
Large language modelsorpus datasetModel evaluation
李晓瑛,刘辉,唐小利,尹浩,任慧玲,罗妍,吕婷钰,刘宇炀,李心怡,杜晋华.中文医学知识大模型问答语料数据集构建研究[EB/OL].(2023-10-28)[2025-08-18].https://www.biomedrxiv.org.cn/article/doi/bmr.202404.00002.点此复制
评论