Hate Speech Detection on Stormfront -> GHC 94 mins annotation (test)
59.5Source F1Source model
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Source modelBackbone=BERT-Large2021.03 | 59.5 | 41.9 | 17.1 | |
| REMOTE (Rsoft + Cstrict)Backbone=RoBERTa-Base2021.03 | 57.6 | 50.1 | 77.5 | |
| Source modelBackbone=RoBERTa-Base2021.03 | 57.4 | 39.6 | 43.8 | |
| REMOTE (Rsoft)Backbone=RoBERTa-Base2021.03 | 57 | 40.9 | 36.7 | |
| Fine-tune (Csample)Backbone=RoBERTa-Base2021.03 | 56.7 | 40.3 | 20.7 | |
| REMOTE (Rsoft)Backbone=BERT-Large2021.03 | 56.4 | 45.7 | 0.5 | |
| Fine-tune (Call)Backbone=RoBERTa-Base2021.03 | 52.2 | 50.5 | 294 | |
| L2-reg. (Csample)Backbone=BERT-Large2021.03 | 49.9 | 45.3 | 12.2 | |
| Fine-tune (Csample)Backbone=BERT-Large2021.03 | 49.6 | 45 | 24.3 | |
| REMOTE (Rsoft + Cstrict)Backbone=BERT-Large2021.03 | 49 | 52.2 | 10 | |
| L2-reg. (Csample)Backbone=RoBERTa-Base2021.03 | 48.8 | 41.7 | 46.2 | |
| Distillation (Csample)Backbone=RoBERTa-Base2021.03 | 48.4 | 40.9 | 28.1 | |
| Distillation (Csample)Backbone=BERT-Large2021.03 | 46.5 | 48.8 | 23.4 | |
| Fine-tune (Call)Backbone=BERT-Large2021.03 | 46 | 53.8 | 142.3 |