YouFake: A Novel Multi-Modal Dataset for Fake News Classification

With the widespread use of social media platforms (such as YouTube), fake news has become a growing concern. Identifying fake news is a challenging task, given the volume of content, the lack of fact-checking, and the disguising of fake news as true news. This study aims to provide a multi-modal fake news dataset that contains both images and texts collected from famous YouTube channels. We labeled our data into 2-way (True, False) and 6-way classes based on categories of fake news such as misleading content, manipulated content, satire/parody, and false connection according to text, thumbnail images, and content provided in videos. In addition, different transfer learning-based models are used to classify fake news using multi-modal data.