LIAR++ 和 FullFact
收藏arXiv2023-08-29 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2308.15202v1
下载链接
链接失效反馈官方服务：
资源简介：
本研究使用了两个不同风格和结构的数据集：LIAR++和FullFact。LIAR++是从POLITIFACT网站收集的政治主题文章，包含6451个三元组（声明、裁决、文章）；FullFact则覆盖更广泛的主题，如健康、经济等，包含1838个三元组。这两个数据集均用于评估自动化事实检查解释生成的泛化能力。LIAR++保留了完整的裁决文本，而FullFact的裁决总是作为网页的独立元素存在。这些数据集的创建旨在通过不同的文本处理方法，如提取和抽象摘要，来生成事实检查的解释，从而帮助自动化事实检查过程。
This study employs two datasets with distinct styles and structures: LIAR++ and FullFact. LIAR++ is a collection of political-themed articles scraped from the POLITIFACT website, containing 6451 triples (statement, verdict, article); FullFact covers a broader range of topics such as health, economics and others, and includes 1838 triples. Both datasets are used to evaluate the generalization capability of automated fact-checking explanation generation. LIAR++ retains the full verdict text, while the verdict in FullFact always exists as a standalone element on web pages. These datasets were created to generate fact-checking explanations via diverse text processing methods including extractive and abstractive summarization, so as to facilitate automated fact-checking processes.
提供机构：
意大利国家研究委员会Bruno Kessler基金会
创建时间：
2023-08-29
