CANDYSET
收藏资源简介:
CANDYSET是一个大规模的中文数据集,旨在系统地评估大型语言模型(LLMs)在事实核查方面的能力和局限性。该数据集由大约20,000个真实新闻和虚假新闻实例组成,涵盖了多个领域,如政治、文化、科学、健康、社会和灾难。数据集还包含了4,891个手动注释的LLM生成的事实核查解释,以及大约7,000个人类研究样本。CANDYSET数据集通过严格的来源控制和人工注释过程,确保了数据的质量和可靠性。该数据集可用于研究LLMs在事实核查方面的缺陷,并探索LLMs在实际场景中的应用潜力。
CANDYSET is a large-scale Chinese dataset designed to systematically evaluate the capabilities and limitations of large language models (LLMs) in fact-checking. It consists of approximately 20,000 real and fake news instances spanning multiple domains including politics, culture, science, health, society and disasters. The dataset also contains 4,891 manually annotated fact-checking explanations generated by LLMs, as well as around 7,000 human research samples. The CANDYSET dataset ensures data quality and reliability through strict source control and manual annotation processes. This dataset can be used to study the shortcomings of LLMs in fact-checking and explore their application potential in real-world scenarios.




