We utilized a pre-trained model to classify Arabic text. After conducting extensive research, we found that MarBERT was the best model for classifying Arabic offensive tweets. It focuses on dialectal Arabic (DA) and Modern Standard Arabic (MSA). The competition involves two shared sub-tasks: detecting whether a tweet is offensive or not; and detecting whether a tweet contains hate speech or not. It detected offensive sentences with 84.9% accuracy and F1-Score of 83.5%, and hate speech with 93.4% accuracy and F1-Score of 80.4%.
https://sites.google.com/view/arabichate2022/home#h.ccedvheaxpk2
