Rationale Extraction on Movies
82.9AccuracyGLASS
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GLASSModel=DeBERTa-v32026.06 | 82.9 | 0.954 | 0.962 | 41.8 | |
| Direct Prompting (DP)Model=Llama3 80B2026.06 | 81.4 | 0.907 | 0.841 | 24.4 | |
| Counterfactual-Sequential (CFS)Model=Llama3 80B2026.06 | 81.4 | 0.889 | 0.795 | 24.3 | |
| Counterfactual-Parallel (CFP)Model=Llama3 80B2026.06 | 79.9 | 0.9 | 0.845 | 21.6 | |
| Counterfactual-Sequential (CFS)Model=GPT-4o2026.06 | 76.4 | 0.913 | 0.864 | 23.6 | |
| Direct Prompting (DP)Model=GPT-4o2026.06 | 74.9 | 0.916 | 0.896 | 24.5 | |
| Counterfactual-Parallel (CFP)Model=GPT-4o2026.06 | 53.3 | 0.809 | 0.705 | 22.4 |