Hate Speech Detection on Davidson (80-20)
90.05F1-score (Class 0)TF-IDF top 75%
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| TF-IDF top 75%Top-k most informative samples=75%, Vocabulary Augmentation=True2025.12 | 90.05 | 90.44 | 89.68 | 97.92 | 98.08 | 98 | 94.03 | 96.67 | |
| TF-IDF top 80%Top-k most informative samples=80%, Vocabulary Augmentation=True2025.12 | 89.8 | 86.79 | 93.04 | 98.57 | 97.14 | 97.85 | 93.83 | 96.5 | |
| TF-IDF top 60%Top-k most informative samples=60%, Vocabulary Augmentation=True2025.12 | 89.8 | 86.96 | 92.08 | 98.38 | 97.14 | 97.85 | 93.62 | 96.39 | |
| TF-IDF top 65%Top-k most informative samples=65%, Vocabulary Augmentation=True2025.12 | 89.65 | 86.92 | 92.56 | 98.48 | 97.19 | 97.83 | 93.74 | 96.45 | |
| TF-IDF top 70%Top-k most informative samples=70%, Vocabulary Augmentation=True2025.12 | 89.42 | 87.57 | 91.36 | 98.24 | 97.38 | 97.81 | 93.62 | 96.4 | |
| Original DavidsonModel=Vanilla BERT, Filtering=None2025.12 | 89.01 | 88.12 | 89.92 | 97.95 | 97.55 | 97.75 | 93.38 | 96.28 | |
| BERT+ Deep Model2025.12 | 89 | — | — | 96 | 96 | 98 | — | 96.2 | |
| TF-IDF top 50%Top-k most informative samples=50%, Vocabulary Augmentation=True2025.12 | 87.48 | 89.05 | 85.95 | 97.18 | 97.87 | 97.52 | 92.5 | 95.84 | |
| HurtBERT2025.12 | — | — | — | — | — | — | 90.8 | — |