Hate Speech Detection on HatEval -> GHC 80 mins annotation (test)
63.7Source F1Source model
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Source modelBackbone=BERT-Large2021.03 | 63.7 | 31.4 | 124.3 | |
| Distillation (Csample)Backbone=BERT-Large2021.03 | 63.1 | 43.3 | 132.7 | |
| Source modelBackbone=RoBERTa-Base2021.03 | 62.7 | 30.9 | 61.6 | |
| L2-reg. (Csample)Backbone=RoBERTa-Base2021.03 | 62.4 | 42.2 | 292.7 | |
| REMOTE (Rsoft + Cstrict)Backbone=BERT-Large2021.03 | 62 | 46.1 | 15.3 | |
| Distillation (Csample)Backbone=RoBERTa-Base2021.03 | 61.8 | 42.1 | 152.5 | |
| REMOTE (Rsoft)Backbone=BERT-Large2021.03 | 61.5 | 37.5 | 55.5 | |
| Fine-tune (Csample)Backbone=RoBERTa-Base2021.03 | 61.3 | 40.8 | 185.2 | |
| REMOTE (Rsoft)Backbone=RoBERTa-Base2021.03 | 61.1 | 40.5 | 15.9 | |
| Fine-tune (Csample)Backbone=BERT-Large2021.03 | 60.8 | 40.7 | 175.1 | |
| L2-reg. (Csample)Backbone=BERT-Large2021.03 | 58.1 | 41.8 | 102.2 | |
| REMOTE (Rsoft + Cstrict)Backbone=RoBERTa-Base2021.03 | 57.5 | 44.7 | 97.8 | |
| Fine-tune (Call)Backbone=RoBERTa-Base2021.03 | 51.4 | 50.6 | 263.2 | |
| Fine-tune (Call)Backbone=BERT-Large2021.03 | 51.3 | 52.5 | 98 |