Tasks/Hate Speech Detection

Hate Speech Detection

ঘৃণাত্মক বক্তব্য শনাক্তকরণ

Identifying hateful, offensive, and abusive Bangla content online.

8 datasets·87 papers
DatasetSizeLicenseYearSource paperLink
BanTH
Multi-label transliterated (romanized) Bangla hate speech dataset from YouTube across 7 target classes.
37,300 samplesMIT2025BanTH (Findings NAACL 2025)
✓ verified 2026-07-19
ALERT
Bangla religious-aggression texts with English translations, labelled hate speech, vandalism, atrocity, or no aggression.
4,027 textsCC BY 4.02025ALERT (Data in Brief 2025)
✓ verified 2026-07-26
Bangla-ToCo
Context-aware toxic-comment corpus retaining news metadata and neighbouring comments, with balanced toxic/non-toxic labels.
1,004 commentsCC BY 4.02025Bangla-ToCo (Data in Brief 2025)
✓ verified 2026-07-26
BanHate
YouTube comments annotated for binary hate, hate-category, and target-group labels.
19,203 commentsMIT2025BanHate (BLP 2025)
✓ verified 2026-08-23
TB-OLID
Offensive language in transliterated (Banglish) text.
5,000 commentsCC BY 4.02023Transliterated offensive language ID
✓ verified 2026-03-08
OffMix-3L
Bangla-English-Hindi code-mixed offensive language identification test set.
1,001 instancesAGPL-3.02023OffMix-3L (SocialNLP 2023)
✓ verified 2026-07-19
BD-SHS
Hierarchically annotated hate speech: target, type, severity.
50,281 commentsCC BY-NC 4.02022BD-SHS (LREC 2022)
✓ verified 2026-06-05
Bengali Hate Speech
v1.0 release with one label per statement: geopolitical, religious, personal, gender abusive, or political hate. Repository license: MIT.
3,418 labelled statementsMIT2020Bengali Hate Speech v1.0 (Karim et al., 2020)
✓ verified 2026-09-08