MFND (Multimodal Fake News Detection) Dataset
收藏资源简介:
MFND数据集是一个大规模且多样化的多模态虚假新闻检测数据集,包含11种操控类型,旨在检测和定位高度真实的虚假新闻。数据集基于原始VisualNews数据集构建,包含约20万个真实社会新闻的图像-文本对。MFND数据集包含了关键词和情感反转、摘要归纳、关键词替换等操控技术。数据集分为训练集、测试集和验证集,分别包含95k、15k和15k个样本。该数据集旨在帮助研究者开发更有效的虚假新闻检测模型,以应对当前日益复杂的虚假新闻问题。
The MFND dataset is a large-scale, diverse multimodal fake news detection dataset encompassing 11 types of manipulation, designed to detect and locate highly realistic fake news. Constructed upon the original VisualNews dataset, it contains approximately 200,000 image-text pairs from real social news. The MFND dataset incorporates manipulation techniques including keyword and sentiment reversal, summary induction, and keyword replacement. It is split into training, test, and validation sets, with 95k, 15k, and 15k samples respectively. This dataset is intended to help researchers develop more effective fake news detection models to address the increasingly complex issue of fake news in the current context.




