Multitask Fake News Detection in Arabic Language using AraELECTRA model: COVID-19 Case Study

In the context of disaster management, the rapid dissemination of accurate information is crucial. Social media platforms, while instrumental in spreading news, also facilitate the circulation of unverified and potentially harmful information, particularly during health crises like the COVID-19 pandemic.To address this challenge, we developed an automated method to detect fake information in Arabic-language tweets. Our approach involved two phases: first, we utilized deep learning models with Word2Vec embeddings and several pretrained transformer models to identify tweets containing fake information. AraELECTRA emerged as the most effective model in this task. Leveraging this model, we then implemented a multitask learning approach to detect both fake information and tweets discussing treatments or vaccines. This dual detection mechanism is essential in disaster management, as it helps prevent the spread of dangerous misinformation that can exacerbate public health crises. Our model achieved an overall accuracy of 98.14%, demonstrating its effectiveness for multitask challenges in the Arabic language.