MAFIA-Net: Multimodal Arabic Fake-news Identification via Hybrid Attention Networks

This repository provides a curated dataset for Arabic multimodal (image + text) fake news detection. It includes tweet IDs, binary labels (Fake/Real), and source dataset identifiers, facilitating research on multimodal misinformation in Arabic social media. This dataset is formally introduced in the publication “MAFIA-Net: Multimodal Arabic Fake-news Identification via Hybrid Attention Networks”, will be published in "Array" Journal, where detailed information about the data collection process, annotation methodology, and experimental use is provided. Data Collection: Arabic tweets were collected from multiple misinformation-related sources, including COVID-19 content, rumors, and fake news datasets. Data acquisition was performed using the Twitter API. Labels were inherited from existing Arabic unimodal annotated datasets, enabling scalable multimodal annotation with minimal manual effort. Text and media content can be programmatically retrieved using the provided tweet IDs. 📝 Linguistic Statistics: Class Tweets Total Words Unique Words Avg. Words / Tweet Real 3777 71,548 19,694 18.94 Fake 1361 30,474 10,158 22.39 Total 5138 102,022 25,431 19.86