Interpretation Applicability Classification on EAF human-consensus set (test)
87AccuracyTransformer
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TransformerSupervision Regime=Human-supervised, Backbone=RoBERTa-base2026.01 | 87 | 87 | 87 | |
| Logistic RegressionSupervision Regime=Human-supervised, Features=TF–IDF2026.01 | 80 | 80 | 80 | |
| TransformerSupervision Regime=Autonomous-supervised (GPT labels), Backbone=RoBERTa-base2026.01 | 78 | 77 | 77 | |
| Linear SVMSupervision Regime=Human-supervised, Features=TF–IDF2026.01 | 77 | 77 | 77 | |
| o1Mode=Zero-Shot2026.01 | 62 | 53 | 54 | |
| Always ApplicableType=Heuristic2026.01 | 53 | 35 | 37 | |
| Always Not ApplicableType=Heuristic2026.01 | 47 | 32 | 30 | |
| RandomType=Baseline2026.01 | 44 | 43 | 44 |