Multimodal Arabic Rumors Detection

Recently, the use of social media platforms has increased with ease and fast accessibility, making such a place rumor proliferation owing to lack posting constraints content authentication. Therefore, there is need leverage artificial intelligence techniques detect rumors on prevent their adverse effects society individuals. Most existing works that in Arabic target textual features tweet content. Nevertheless, tweets contain different types content, as (text, images, videos, URLs), visual play an essential role diffusion. This study proposes detection model Twitter using image through two multimodal fusion: early late fusion. In addition, we leveraged transfer learning pre-trained language vision models. Different experiments were conducted select best feature extractors for building model. MARBERTv2 was used extractor, whereas ensemble VGG-19 ResNet50 extractor build Subsequently, models single baseline compare results those Finally, experimental demonstrate effectiveness tasks compared