GM07/med_hal
收藏Hugging Face2025-04-11 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/GM07/med_hal
下载链接
链接失效反馈官方服务：
资源简介：
MedHal是一个用于评估临床环境中虚构内容检测的基准数据集。该数据集汇集了四个任务（问答、自然语言推理、摘要和信息提取），围绕多个临床文档（临床试验、临床笔记、医学问题和科学论文）。数据集的目的是让LLM模型评估一个陈述是否真实，如果陈述中的所有信息都能得到一般医学知识或所提供上下文的支持，则模型应回答YES。
MedHal is an evaluation dataset for detecting hallucinated content in clinical settings. The dataset comprises four tasks (QA, NLI, Summarization, Information Extraction) centered around multiple clinical documents (clinical trials, clinical notes, medical questions, and scientific papers). The goal is to have LLMs evaluate whether a statement is factual or not, with a YES answer required if all information mentioned in the statement is backed up by general medical knowledge or the provided context.
提供机构：
GM07
