HateBRXplain: The Evaluation Benchmark for Explainable Hate Speech Detection in Brazilian Portuguese

HateBR is the first large-scale evaluation benchmark for explainable hate speech detection in Brazilian Portuguese. It contains 7,000 documents annotated with a binary classification (offensive vs. non-offensive comments), where the offensive comments (3,500) include human-annotated rationales. The rationales were annotated by two different annotators.

Paper:

HateBRXplain: A Benchmark Dataset with Human-Annotated Rationales for Explainable Hate Speech Detection in Brazilian Portuguese
Isadora Salles, Francielle Vargas, Fabrício Benevenuto
31st International Conference on Computational Linguistics (COLING 2025)
Abu Dhabi, UAE. pp. 6659-6669. 2025. https://aclanthology.org/2025.coling-main.446/