Semantic Textual Similarity on STS-B (Accuracy/Bias evaluation)
0.595AccuracyEAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| EARBackbone=RoBERTa-base2024.06 | 0.595 | 0.333 | |
| MABELBackbone=RoBERTa-base2024.06 | 0.591 | 0.304 | |
| Vanilla-tuningBackbone=RoBERTa-base2024.06 | 0.578 | 0.33 | |
| MABELBackbone=BERT-base2024.06 | 0.57 | 0.181 | |
| Debiased-tuningBackbone=RoBERTa-base2024.06 | 0.518 | 0.314 | |
| EARBackbone=BERT-base2024.06 | 0.509 | 0.233 | |
| Vanilla-tuningBackbone=BERT-base2024.06 | 0.507 | 0.197 | |
| ProSocialTuningBackbone=RoBERTa-base2024.06 | 0.494 | 0.28 | |
| Debiased-tuningBackbone=BERT-base2024.06 | 0.473 | 0.184 | |
| ProSocialTuningBackbone=BERT-base2024.06 | 0.46 | 0.169 |