Claim Detection
89.55F1 (Task)STL Classification (CLS)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| STL Classification (CLS)Learning Paradigm=Single-Task Learning, Backbone=Qwen3-4B, Configuration=Classification Head2026.01 | 89.55 | 95.17 | 92.36 | 93.26 | |
| MTL CLSLearning Paradigm=Multi-Task Learning, Backbone=Qwen3-4B, Configuration=Classification Head2026.01 | 88.37 | 94.06 | 91.21 | 92.12 | |
| MTL Instruction TuningLearning Paradigm=Multi-Task Learning, Backbone=Qwen3-4B, Configuration=Instruction Tuning2026.01 | 85.4 | 93.7 | 89.6 | 90.9 | |
| STL Instruction TuningLearning Paradigm=Single-Task Learning, Backbone=Qwen3-4B, Configuration=Instruction Tuning2026.01 | 81.3 | 92.51 | 86.92 | 88.7 | |
| Zero-shotMode=Zero-shot2026.01 | 80.4 | 71.3 | 75.9 | 77.3 | |
| XLM-RoBERTa-LargeBackbone=XLM-RoBERTa-Large2026.01 | 80 | 90.6 | 85.3 | 86.99 | |
| STL Causal LMLearning Paradigm=Single-Task Learning, Backbone=Qwen3-4B, Configuration=Causal Language Model2026.01 | 78.18 | 86.51 | 82.35 | 83.68 | |
| MTL CLMLearning Paradigm=Multi-Task Learning, Backbone=Qwen3-4B, Configuration=Causal Language Model2026.01 | 70.58 | 86.11 | 90.71 | 80.83 | |
| Few-shot ICLMode=Few-shot In-Context Learning2026.01 | 69.3 | 82.8 | 76.1 | 78.2 |