ArAIEval Shared Task: Propagandistic Techniques Detection in Unimodal and Multimodal Arabic Content

The second edition of the ArAIEval shared task was successfully presented at the ArabicNLP 2024 conference, co-located with ACL 2024. This year’s task focused on the critical challenge of detecting propagandistic techniques in Arabic content, offering two distinct challenges: identifying persuasion techniques in textual spans from tweets and news articles, and distinguishing between propagandistic and non-propagandistic memes.

A total of 14 teams participated in the final evaluation, with 6 teams participated in the first task and 9 teams participated with the second. The results demonstrated the efficacy of fine-tuning transformer models, such as AraBERT, which formed the backbone of many participating systems. In addition to highlighting the task setup, the organizers released all datasets and evaluation scripts to the research community, aiming to encourage further exploration in this vital area of Arabic language processing.

The shared task attracted considerable interest, culminating in 11 teams submitting detailed system description papers, further contributing to the advancement of research on propagandistic content detection in Arabic.