Argument Relation Identification and Classification (ARIC) on UKP Student Essays (test)
58.5Macro F1Proposed
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ProposedEvaluation Protocol=Inference-only, Prompting Strategy=multi-agent debate, Confidence Threshold τ=0.702026.06 | 58.5 | 67.2 | 63 | 42.7 | 69.8 | |
| Baseline CEvaluation Protocol=Inference-only, Prompting Strategy=Smart Reasoning2026.06 | 57.8 | 66.6 | 63.5 | 41.2 | 68.8 | |
| ProposedEvaluation Protocol=Inference-only, Prompting Strategy=multi-agent debate, Configuration=full debate2026.06 | 56.1 | 66.4 | 60.9 | 37.8 | 69.7 | |
| Baseline BEvaluation Protocol=Inference-only, Prompting Strategy=Chain of Thought (CoT)2026.06 | 56 | 68.4 | 63.6 | 32.7 | 71.6 | |
| Baseline AEvaluation Protocol=Inference-only, Prompting Strategy=Vanilla2026.06 | 54.9 | 65.8 | 62.3 | 34 | 68.3 | |
| RoBERTa-largeEvaluation Protocol=Supervised2026.06 | 52.2 | 72.1 | 61 | 16.7 | 78.8 | |
| RoBERTa-baseEvaluation Protocol=Supervised2026.06 | 47.3 | 70.7 | 55.2 | 7.1 | 79.7 |