CFEVER

该数据集名为CFEVER，它遵循FEVER的标注方法构建而成，该方法涉及基于维基百科数据生成主张，并对这些主张标注为“支持”、“反驳”或“信息不足”。这些主张是由母语为中文的说话者生成的，数据集中包含了需要来自多个页面的证据的各种主张类型。该数据集包含了来自2022年12月版中文维基百科的1,187,751个页面，主张是基于访问量最高的页面生成的。其任务是为事实核实生成和标注主张。

This dataset is named CFEVER, which is constructed following the annotation methodology of the FEVER dataset. This methodology entails generating claims based on Wikipedia data and annotating these claims with three standard labels: "Supports", "Refutes", or "Not Enough Information". These claims are generated by native Chinese speakers, and the dataset encompasses diverse claim types that necessitate evidence from multiple Wikipedia pages. The dataset includes 1,187,751 pages sourced from the December 2022 edition of the Chinese Wikipedia, with the claims built upon the most frequently visited pages within this corpus. The core task of this dataset is to generate and annotate claims for fact verification.

搜集汇总

背景与挑战

背景概述

FEVER是一个用于事实提取与验证的大规模数据集，旨在支持自然语言处理中的事实核查任务。该数据集包含注释平台代码和基线模型，源自NAACL 2018论文，并用于相关共享任务，以促进事实验证技术的研究与发展。