MedEthicsQA
收藏资源简介:
MedEthicsQA是一个全面的评估医疗伦理的大型语言模型(MedLLM)的问答基准。该数据集包含5623个多项选择题和5351个开放式问题,用于评估MedLLM的伦理安全性。数据集整合了全球医疗伦理标准,包括广泛使用的医疗数据集、权威题库和来自PubMed文献的场景。通过多阶段过滤和多方面专家验证的严格质量控制,确保数据集的可靠性,错误率低至2.72%。该数据集旨在解决现有医疗语言模型在医疗伦理方面的不足,确保其在提供安全临床辅助方面的道德准则遵守。
MedEthicsQA is a comprehensive question answering benchmark for evaluating large language models dedicated to medical ethics (MedLLMs). This dataset comprises 5,623 multiple-choice questions and 5,351 open-ended questions, designed to assess the ethical safety of MedLLMs. It integrates global medical ethics standards, incorporating widely adopted medical datasets, authoritative question banks, and scenarios extracted from PubMed literature. Rigorous quality control workflows, including multi-stage filtering and multi-faceted expert validation, have been implemented to guarantee the dataset's reliability, with an error rate as low as 2.72%. This benchmark aims to address the shortcomings of existing medical language models in medical ethics, ensuring their adherence to ethical guidelines when delivering safe clinical assistance.
MedEthicsQA数据集概述
数据集简介
- 名称:MedEthicsQA
- 类型:医疗伦理问答评估基准
- 用途:用于评估大型语言模型(LLMs)在医疗伦理领域的表现
数据集组成
主要文件
-
MedEthicsQA_MCQ.json- 格式:JSON (UTF-8编码)
- 内容:多项选择题子集
-
MedEthicsQA_open.zip- 格式:压缩文件
- 内容:开放式问题子集
-
taxonomy.json- 内容:提出的分层分类法结构
评估脚本
-
calculate_acc.py- 功能:计算多项选择题的准确率
-
calculate_ratings.py- 功能:计算开放式问题的评分
-
evaluate_MCQ_api.py- 功能:多项选择题的推理文件
-
evaluate_open-ended_api.py- 功能:开放式问题的推理文件
-
open-ended-LLM-as-Judge.py- 功能:评估LLM对开放式问题的回答与真实答案的匹配度
引用信息
bibtex @misc{wei2025medethicsqacomprehensivequestionanswering, title={MedEthicsQA: A Comprehensive Question Answering Benchmark for Medical Ethics Evaluation of LLMs}, author={Jianhui Wei and Zijie Meng and Zikai Xiao and Tianxiang Hu and Yang Feng and Zhijie Zhou and Jian Wu and Zuozhu Liu}, year={2025}, eprint={2506.22808}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2506.22808}, }

- 1MedEthicsQA: A Comprehensive Question Answering Benchmark for Medical Ethics Evaluation of LLMs浙江大学和Angelalign Technology Inc. · 2025年



