RAWFC

据我们所知，没有可用于此任务的原始报告的公共数据集。因此，我们收集了两个可解释的数据集，即RAWFC和LIAR-RAW，分别指的是黄金标签的两个不同的事实检查站点 (即Snopes3和Politifact4)
对于RAWFC，我们通过检索索赔关键字从Snopes收集索赔和相关原始报告，从头开始构建它。
为了减轻事实检查报告的依赖性，RAWFC是通过使用原始报告 (从头开始) 构建的，其中黄金标签指的是Snopes。train/val/test集中的每个实例都显示为signle文件。

To the best of our knowledge, there are no public datasets of original reports available for this task. Therefore, we collected two interpretable datasets, RAWFC and LIAR-RAW, which correspond to two distinct fact-checking websites with gold labels, namely Snopes³ and Politifact⁴ respectively. For RAWFC, we constructed it from scratch by retrieving claim keywords from Snopes to collect claims and their related original reports. To mitigate the dependency on fact-checking reports, RAWFC was built using original reports (developed from scratch), where the gold labels are sourced from Snopes. Each instance in the train/val/test split is presented as a single file.

搜集汇总

背景与挑战

背景概述

RAWFC是一个用于事实检查任务的数据集，通过从Snopes收集索赔和相关原始报告构建，旨在减轻对事实检查报告的依赖性。该数据集由吉林大学和香港浸会大学于2022年发布，以单个文件形式组织训练、验证和测试集，黄金标签参考Snopes。