trends.barnas.net

task · first seen 11 Oct 2026

hate speech detection

Also: misogyny detection, offensive language detection

Automatically identifying hateful or abusive language directed at groups or individuals in text.

Hate speech detection is the long-standing NLP task of classifying text as hateful, abusive or offensive toward groups or individuals. The recent papers do not introduce a new idea. They continue the steady stream of language-specific datasets and benchmarks for Nastaliq and Roman Urdu, Turkish sarcasm, Persian, Bengali misandry, Brazilian Portuguese and Polish misogyny, and Danish. Others study cross-lingual transfer across 14 languages, evaluate prompting of open-weight LLMs against BERT baselines, and probe LLM classifiers for identity-dependent measurement non-invariance. The common threads are low-resource languages, explainability and using LLMs as classifiers, all within a well-established research area.

The apparent rise is ordinary publication volume in a mature field. The useful developments are incremental: more resources for low-resource languages and closer scrutiny of whether LLM-based moderation classifiers treat different identity groups consistently.

Analyst note (established), 17 Nov 2026.

Mentions per week

0419 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

SHAPsarcasm detection effect on hate speech detectionPersian language NLPlate fusioncross-lingual transfermeasurement non-invariancecounterfactual identity templatesprompt tuningfew-shot promptingprompt engineering