Relevance Labeling on BEIR RobustQA
94.1Class-wise Recall (Irrelevance)LARA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LARAEscalation Ratio=50.0%2026.02 | 94.1 | 98.4 | 96.3 | |
| DREAMEscalation Ratio=3.5%2026.02 | 91.9 | 98.4 | 95.2 | |
| Human-OnlyEscalation Ratio=100.0%2026.02 | 89.9 | 97.8 | 93.8 | |
| LARAEscalation Ratio=25.0%2026.02 | 80.2 | 95.3 | 87.8 | |
| LARAEscalation Ratio=12.5%2026.02 | 76.1 | 91.6 | 83.9 | |
| LARAEscalation Ratio=3.5%2026.02 | 74.5 | 89.6 | 82.1 | |
| LLMJudgeEscalation Ratio=0.0%2026.02 | 50.2 | 97.5 | 73.9 |