AraTox: A Multi-Dialect, Multi-Label Arabic Dataset for Toxicity Detection

AraTox is a multi-dialect, multi-label Arabic dataset for toxicity detection. It contains annotated Arabic text representing multiple varieties, including Gulf, Levantine, Nile Basin, North African, Yemeni, and Modern Standard Arabic (MSA). The dataset is designed to support research in toxic language detection, multi-label classification, and the benchmarking of Arabic NLP models across dialects. This dataset accompanies the following publication: Aratox: A multi-dialect, multi-label Arabic dataset and model benchmark for toxicity detection If you use this dataset, please cite: Alshargi, F., Abulohoom, A., Yagi, S., Jabr, F., Lulu, L., & Elnagar, A. (2026). Aratox: A multi-dialect, multi-label Arabic dataset and model benchmark for toxicity detection. Language Resources & Evaluation, 60, 39. doi.org