A Framework for Sexism Detection on Social Media via ByT5 and TabNet

Hateful and offensive content on social media platforms particularly content directed towards a specific gender is a great impediment towards equality, diversity and inclusion. Social media platforms are facing increasing pressure to work towards regulation of such content; and this has directed researchers in text mining to work towards hate speech identification algorithms. One such attempt is sexism detection for which mostly transformer-based text methods have been proposed. We propose a combination of byte-level model ByT5 with tabular modeling via TabNet that has at its core an ability to take into account platform and language aspects of the challenging task of sexism detection. Despite not performing well in the sexism detection task for IberLEF our approach shows promise for future research in the area.