Meta-review generation on PeerSum (test)
97CoverageAspect-aware decomposition
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Aspect-aware decompositionBackbone LLM=Llama-3.1-70B-Instruct2025.01 | 97 | 76 | 76 | 9 | 22.58 | |
| Sentiment CoTBackbone LLM=GPT-4o2025.01 | 96 | 75 | 72 | 8 | 23.47 | |
| Aspect-aware decompositionBackbone LLM=GPT-4o2025.01 | 95 | 76 | 68 | 6 | 20.78 | |
| Aspect-aware decompositionBackbone LLM=Llama-3.1-8B-Instruct2025.01 | 90 | 66 | 71 | 7 | 21.12 | |
| FT-Llama 8BBackbone LLM=Llama-3.1-8B-Instruct, Protocol=Fine-tuning2025.01 | 87 | 60 | 33 | 6 | 20.6 | |
| Chunk-wise decompositionBackbone LLM=Llama-3.1-70B-Instruct2025.01 | 84 | 72 | 65 | 6 | 21.8 | |
| Chunk-wise decompositionBackbone LLM=Llama-3.1-8B-Instruct2025.01 | 79 | 65 | 65 | 3 | 21.19 | |
| Naive aspect-aware promptingBackbone LLM=Llama-3.1-8B-Instruct2025.01 | 72 | 62 | 70 | 6 | 16.93 | |
| Naive aspect-aware promptingBackbone LLM=Llama-3.1-70B-Instruct2025.01 | 72 | 62 | 70 | 7 | 16.82 | |
| Automatic decompositionBackbone LLM=Llama-3.1-70B-Instruct2025.01 | 59 | 31 | 51 | 3 | 12 | |
| Automatic decompositionBackbone LLM=Llama-3.1-8B-Instruct2025.01 | 58 | 20 | 36 | 3 | 11.98 |