Court views generation on US Court Views (test)
63.3ROUGE-1D3LM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| D3LM2024.06 | 63.3 | 53.1 | 59.2 | 38.7 | 31.7 | 26.9 | |
| GPT4.0-STInteraction=single-turn2024.06 | 60.3 | 42.7 | 57.1 | 27.2 | 18.2 | 14.1 | |
| D3LMw/oRFVariant=without reinforcement learning2024.06 | 54.1 | 31.9 | 54.1 | 39.6 | 29.9 | 25.1 | |
| D3LMw/oPUVariant=without positive-unlabeled learning2024.06 | 51.5 | 31.3 | 51.5 | 35.1 | 25.8 | 21.5 | |
| D3LMw/oLMRCVariant=without language model reading comprehension2024.06 | 50 | 27.9 | 44.1 | 35.6 | 27.4 | 22.7 | |
| FT-Llama2-13B-STTraining=Fine-tuned, Interaction=single-turn2024.06 | 39.7 | 26.5 | 41.8 | 24.3 | 17.4 | 13.2 | |
| Llama2-13B-STInteraction=single-turn2024.06 | 37.5 | 24.3 | 37.5 | 20.8 | 14.6 | 11.4 | |
| AC-NLG2024.06 | 35.8 | 18.5 | 32.7 | 23.6 | 16.2 | 12.7 | |
| BM25type=retrieval baseline2024.06 | 30.8 | 16.1 | 28.2 | 20.1 | 13.9 | 10.9 |