FACTIFY

该数据集名为Factify，是一个用于多模态事实核实的数据库，其中包含了图像、文本主张、参考文本文件以及相应的图像。每个样本都包括主张图像、主张文本、主张光学字符识别（OCR）内容、文件图像、文件文本、文件OCR内容以及类别标签。该数据集共分为五个类别：支持多模态、支持文本、多模态不足、文本不足和反驳。在规模上，该数据集包含了训练集35,000个样本、验证集7,500个样本以及测试集7,500个样本。其任务是进行多模态事实核实。

This dataset is named Factify, a specialized database for multimodal fact verification. Each sample in the dataset includes claim image, claim text, claim optical character recognition (OCR) content, document image, document text, document OCR content, and category label. The dataset is divided into five categories: Multimodal-Supported, Text-Supported, Multimodal-Insufficient, Text-Insufficient, and Refuted. In terms of scale, it contains 35,000 training samples, 7,500 validation samples, and 7,500 test samples. The core task of this dataset is multimodal fact verification.