Automatic Fact CheckingThrough Multimodal Knowledge Bases

社交媒体平台目前是信息和意见形成内容的主要来源。然而，由于其促进用户之间互动的内在本质，这些平台充斥着有争议的，耸人听闻的，往往是故意虚假的谣言，可能会造成经济，政治和社会危害。多年来，自动揭穿虚假谣言的任务有了显着改善，许多作品提出了不同的方法来解决这个问题。然而，大多数谣言检测系统仅仅依赖于文本或图像/视频数据，而没有探索谣言的多模态方面。在相关任务中证明有价值的一种方法是使用知识库。因此，该项目建议学习和应用多模态知识库来揭穿虚假谣言或支持真实和经过验证的信息。

Social media platforms are currently a primary source of information and opinion-formingcontent. However, due to their intrinsic nature of promoting engagement between its users,these platforms are rife with controversial, sensationalist, and often intentionally false rumors that can potentially cause economic, political, and social harms. The task of automatically debunking false rumors has significantly improved over the years, with many works proposing different approaches to solve the problem. Nevertheless, most of the rumor detection system relies solely on textual or on image/video data separately, without exploring the multimodality aspect of rumors. One approach that has proven value in related tasks is the use of knowledge bases. Therefore, this project proposes learning and applying a multimodal knowledge base to debunk false rumors or support true and verified information.