Document-level Information Extraction on MUC English 4
58.32MUC ScoreTHINKTWICE
Evaluation Results
| Method | Links | |
|---|---|---|
| THINKTWICEBackbone=Qwen 3, Selector=oracle, Protocol=Supervised2026.01 | 58.32 | |
| THINKTWICEBackbone=Llama R1, Selector=oracle, Protocol=Supervised2026.01 | 56.6 | |
| THINKTWICEBackbone=Qwen 3, Selector=Reward, Protocol=Supervised2026.01 | 42.13 | |
| THINKTWICEBackbone=Llama R1, Selector=Reward, Protocol=Supervised2026.01 | 41.99 | |
| THINKTWICEBackbone=Qwen 3, Selector=F1 Voting, Protocol=Supervised2026.01 | 41.4 | |
| THINKTWICEBackbone=Llama R1, Selector=F1 Voting, Protocol=Supervised2026.01 | 37.24 | |
| THINKTWICEBackbone=Qwen 3, Selector=Majority, Protocol=Supervised2026.01 | 36.6 | |
| GreedyBackbone=Qwen 3, Selector=Greedy decoding, Protocol=Supervised2026.01 | 35.97 | |
| ITERXBackbone=T5-enc large, Selector=None, Protocol=Supervised2026.01 | 35.2 | |
| GTTBackbone=BERT base, Selector=None, Protocol=Supervised2026.01 | 32.3 | |
| GreedyBackbone=Llama R1, Selector=Greedy decoding, Protocol=Supervised2026.01 | 28.52 | |
| TEMPGENBackbone=BART large, Selector=None, Protocol=Supervised2026.01 | 28.3 | |
| THINKTWICEBackbone=Llama R1, Selector=Majority, Protocol=Supervised2026.01 | 25.06 |