FACTIFY5WQA

该数据集名为FACTIFY5WQA，专为事实核实而设计，侧重于根据给定证据判断声明的真实性。它包含了利用5W框架从声明中生成的增强型问题。每个样本包含声明、证据、问题、声明答案、证据答案以及指示声明真实性（支持、中性或反驳）的标签。该数据集使用BLEU分数进行评估，规模为15,000个样本，其中10,500个用于训练，2,250个用于验证，另外2,250个用于测试。其任务是进行事实核实。

This dataset, named FACTIFY5WQA, is specifically designed for fact verification tasks, focusing on determining the veracity of claims against provided evidence. It contains enhanced questions generated from claims via the 5W framework. Each sample consists of the claim, evidence, generated question, claim answer, evidence answer, and a label indicating the veracity of the claim (support, neutral, or refuted). The dataset has a total of 15,000 samples, with 10,500 allocated for training, 2,250 for validation, and the remaining 2,250 for testing. It is evaluated using the BLEU score, and its core task is fact verification.