Opinion Summarization on SPACE hotels (test)
1CoverageAspect-aware decomposition
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Aspect-aware decompositionBackbone=GPT-4o2025.01 | 1 | 0.9 | 0.81 | 0.1 | 21.38 | |
| Aspect-aware decompositionBackbone=Llama 70B2025.01 | 0.99 | 0.88 | 0.79 | 0.11 | 23.46 | |
| TCG2025.01 | 0.98 | 0.66 | 0.66 | 0.11 | 22.98 | |
| Aspect-aware decompositionBackbone=Llama 8B2025.01 | 0.97 | 0.81 | 0.7 | 0.1 | 22.05 | |
| Chunk-wise decompositionBackbone=Llama 8B2025.01 | 0.94 | 0.8 | 0.65 | 0.14 | 22.9 | |
| Chunk-wise decompositionBackbone=Llama 70B2025.01 | 0.93 | 0.84 | 0.65 | 0.01 | 22.02 | |
| HIRO-abs2025.01 | 0.87 | 0.62 | 0.83 | 0.24 | 26.5 | |
| Automatic decompositionBackbone=Llama 8B2025.01 | 0.65 | 0.07 | 0.55 | 0.15 | 13.8 | |
| Automatic decompositionBackbone=Llama 70B2025.01 | 0.63 | 0.38 | 0.7 | 0.22 | 10 | |
| Naive aspect-aware promptingBackbone=Llama 8B2025.01 | 0.55 | 0.06 | 0.34 | 0.18 | 10.3 | |
| Naive aspect-aware promptingBackbone=Llama 70B2025.01 | 0.37 | 0.34 | 0.44 | 0.22 | 5 |