ncbi/MedFact-Bench

MedFact-Bench是一个生物医学验证基准，包含五个生物医学验证数据集：SciFact、HealthVer、MedAESQA、PubMedQA-Fact（重新利用的PubMedQA）和BioASQ-Fact（重新利用的BioASQ）。每个实例包括一个声明-源对，标注为三种标签之一：支持（SUPPORT）、信息不足（NEI）或矛盾（CONTRADICT）。数据集特征包括实例的来源数据集、待验证的声明、用于验证的源、真实标签以及Med-V1验证提示的系统部分和用户部分。该数据集旨在用于生物医学声明验证领域的研究，并附有关于其用于医疗决策的免责声明。

MedFact-Bench is a benchmark comprising five biomedical verification datasets: SciFact, HealthVer, MedAESQA, PubMedQA-Fact (re-purposed PubMedQA), and BioASQ-Fact (re-purposed BioASQ). Each instance in the dataset consists of a claim–source pair annotated with one of three labels: SUPPORT, not enough information (NEI), or CONTRADICT. The dataset features include the source dataset of the instance, the claim to be verified, the source used for verification, the ground-truth label, and the system and user parts of the Med-V1 verification prompt. The dataset is intended for research in the field of biomedical claim verification and comes with a disclaimer regarding its use for medical decision-making.