Punitive Damages Identification on Judicial Opinions (test)
77.1AccAgentic LLM + ToD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Agentic LLM + ToDReasoning Strategy=Tree-of-Discourse (ToD) augmented workflow, Components=Discourse-level rhetorical structure2026.01 | 77.1 | 61.9 | 100 | 76.5 | |
| Vanilla LLMReasoning Strategy=Direct2026.01 | 71.4 | 56.5 | 100 | 72.2 | |
| CoTReasoning Strategy=Chain-of-Thought2026.01 | 71.4 | 57.1 | 92.3 | 70.6 | |
| Agentic LLMReasoning Strategy=Agentic (Step-by-step decomposition)2026.01 | 65.7 | 52.2 | 92.3 | 66.7 | |
| Random Baseline2026.01 | 51.7 | 37 | 42.3 | 39.2 |