CHEF中文事实核查数据集

CHEF数据集使用真实世界的中文声明，填补了中文事实核查数据集的空白。CHEF使用搜索引擎返回的文档作为原始证据，更加贴近真实场景，并且使用人类标注返回文档的相关句子作为细粒度的证据，可以用于训练核查系统学会如何搜集证据。CHEF包含了10000条数据，来自中国互联网辟谣平台、腾讯求真等多个企事业平台，涵盖了政治、公卫、社会、科学、文化等领域。

The CHEF dataset, which utilizes real-world Chinese claims, fills the gap in Chinese-language fact-checking datasets. Using documents returned by search engines as original evidence, CHEF better aligns with real-world scenarios. Additionally, it adopts human-annotated relevant sentences from the returned documents as fine-grained evidence, which can be used to train fact-checking systems to master evidence collection skills. The dataset contains 10,000 samples sourced from multiple enterprise and institutional platforms including the China Internet Rumor Debunking Platform and Tencent Truth-Seeking, covering fields such as politics, public health, society, science, and culture.