Table 2. Benchmark Arabic hate speech datasets: sources, reference publications, collection platforms, sizes, and annotation labels.

Dataset/link	Ref./Year	Platform	Size	Labels
Alakrot A.etal dataset	Alakrot, Murray & Nikolov (2018a)	YouTube	15,050	Not offensive, Offensive
Religious Hate Speech Detection dataset	Albadi, Kurdi & Mishra (2018)	Twitter	5,569	Hate, Not hate
MLMA-hate-speech	Ousidhoum et al. (2021a)	Twitter	3,353	Disability, Gender, Religion, Sexual orientation
L-HSAB-First-Arabic-Levantine-Hate Speech-Dataset	Mulki et al. (2019)	Twitter	5,846	Normal, Hate, Abusive
OSACT4 Shared Task on Offensive Language Detection (Subtask A and B)	Hassan et al. (2020)	Twitter	10,000	Task A: OFF, NOT OFF/Task B: HS, NOTH S/
COVID-19-Arabic Tweets-Dataset	Alshalan et al. (2020)	Twitter	975,316 tweet used in Alshalan et al. (2020)	Hate (low, average, high), Non hate
Dataset Hate speech detection in Arabic Twittersphere	Alshalan & Al-Khalifa (2020)	Twitter	9,316	Abusive, hateful, normal
Multi Platforms Offensive Language Dataset (MPOLD)	Chowdhury et al. (2020)	Facebook, Twitter, YouTube	4,000	OFF, NOT OFF
Fine-Grained H.S Detection on Arabic Twitter	Ben Nessir et al. (2022), Shapiro, Khalafallah & Torki (2022)	Twitter	13,000	Subtask A: OFF, NOT OFF/Subtask B: HS, NOT HS/Subtask C: Ideology, Religion, disability, race, Social class, and Gender
Arabic Hate Speech Dataset 2023	Ahmad et al. (2024)	Twitter	403,688	Negative, Neutral, Positive, Very positive