Papers › HateBERT: Retraining BERT for Abusive Language Detection in English

HateBERT: Retraining BERT for Abusive Language Detection in English

23 Oct 2020ACL (WOAH) 2021 8arXiv:2010.12472archive 2025-07-28

Tommaso Caselli, Valerio Basile, Jelena Mitrović, Michael Granitzer

In this paper, we introduce HateBERT, a re-trained BERT model for abusive language detection in English. The model was trained on RAL-E, a large-scale dataset of Reddit comments in English from communities banned for being offensive, abusive, or hateful that we have collected and made available to the public. We present the results of a detailed comparison between a general pre-trained language model and the abuse-inclined version obtained by retraining with posts from the banned communities on three English datasets for offensive, abusive language and hate speech detection tasks. In all datasets, HateBERT outperforms the corresponding general BERT model. We also discuss a battery of experiments comparing the portability of the generic pre-trained language model and its corresponding abusive language-inclined counterpart across the datasets, indicating that portability is affected by compatibility of the annotated phenomena.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

tommasoc80/HateBERT officialmentioned on GitHub report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Abusive LanguageHate Speech DetectionLanguage ModelingLanguage Modelling

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Hate Speech Detection AbusEval HateBERT Macro F1 0.742 #1 of 2 Archive leaderboard report
Hate Speech Detection AbusEval BERT Macro F1 0.724 #2 of 2 Archive leaderboard report
Hate Speech Detection HatEval HateBERT Macro F1 0.494 #1 of 2 Archive leaderboard report
Hate Speech Detection HatEval BERT Macro F1 0.48 #2 of 2 Archive leaderboard report
Hate Speech Detection OffensEval 2019 HateBERT Macro F1 0.805 #1 of 2 Archive leaderboard report
Hate Speech Detection OffensEval 2019 BERT Macro F1 0.803 #2 of 2 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionAttention DropoutBERTDense ConnectionsDropoutLayer NormalizationLinear LayerLinear Warmup With Linear DecayMulti-Head AttentionResidual ConnectionSoftmaxWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections