Neutralized Summarization on NeuS
4.83Arousal (Positive)Pegasus-Multi
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Pegasus-MultiBackbone=Pegasus-large, Training Dataset=Multi-News2025.06 | 4.83 | 35.33 | 13.27 | 21.25 | 31.14 | 14.79 | 59.15 | 2.27 | 7.1 | |
| BART-MultiBackbone=BART-large, Training Dataset=Multi-News2025.06 | 3.46 | 39.6 | 16.26 | 24.57 | 35.05 | 17.6 | 54.47 | 1.64 | 5.1 | |
| GPT-3.5Backbone=gpt-3.5-turbo, Prompting=zero-shot2025.06 | 3.4 | 42.01 | 16.25 | 26.13 | 37.27 | 18.77 | 77.25 | 2.13 | 5.52 | |
| GPT-4Backbone=gpt-4, Prompting=zero-shot2025.06 | 3.37 | 42.36 | 16.49 | 26.3 | 37.31 | 19.04 | 75.86 | 1.97 | 5.34 | |
| GPT-3.5 + one-shotBackbone=gpt-3.5-turbo, Prompting=one-shot2025.06 | 2.52 | 41.95 | 16.77 | 28.13 | 37.39 | 18.33 | 44.29 | 1.57 | 4.08 | |
| LexRankLearning Setting=Unsupervised2025.06 | 2.42 | 42.24 | 18.16 | 26.61 | 36.87 | 17.68 | 63.73 | 1.53 | 3.95 | |
| GPT-4 + graphBackbone=gpt-4, Graph Information=event relation graph2025.06 | 2.11 | 42.61 | 18.67 | 30.82 | 38.18 | 19.09 | 31.77 | 1.49 | 3.6 | |
| GPT-3.5 + graphBackbone=gpt-3.5-turbo, Strategy=chain-of-thought, Graph Information=event relation graph2025.06 | 2.03 | 43.2 | 18.06 | 30.56 | 38.08 | 18.99 | 32.37 | 1.59 | 3.62 | |
| BART-CNNBackbone=BART-large, Training Dataset=CNN/Daily Mail2025.06 | 2.02 | 38.22 | 15.73 | 25.52 | 34.25 | 15.26 | 76.67 | 1.14 | 3.16 | |
| Pegasus-CNNBackbone=Pegasus-large, Training Dataset=CNN/Daily Mail2025.06 | 1.98 | 38.17 | 15.61 | 25.18 | 31.17 | 13.69 | 74.45 | 1.15 | 3.12 | |
| Llama 2Backbone=llama-2-7b-chat-hf, Training Method=LoRA, Learning Setting=Fine-tuned2025.06 | 1.8 | 42.26 | 19.25 | 30.88 | 37.75 | 19.15 | 30.3 | 1.02 | 2.81 | |
| NeuSLearning Setting=Hierarchical abstractive summarization2025.06 | 1.69 | 39.09 | 18.93 | 29.74 | 35.35 | 16.21 | 38.51 | 0.83 | 2.53 | |
| LEDBackbone=led-large-16384, Learning Setting=Fine-tuned2025.06 | 1.59 | 40.3 | 18.63 | 30.24 | 36.26 | 17.3 | 31.97 | 0.86 | 2.45 | |
| Llama-2 + textual graphBackbone=llama-2-7b-chat-hf, Graph Integration=textual graph2025.06 | 1.59 | 43.98 | 20.52 | 32.57 | 39.3 | 20.18 | 28.22 | 0.94 | 2.53 | |
| Bang et al. (2023)Objective=Polarity minimization loss2025.06 | 1.57 | — | — | — | — | — | — | 0.91 | 2.48 | |
| Llama-2 + both (full model)Backbone=llama-2-7b-chat-hf, Graph Integration=textual graph + graph prompt2025.06 | 1.55 | 45.14 | 22.3 | 34.02 | 40.74 | 21.89 | 27.89 | 0.9 | 2.46 | |
| Llama-2 + graph promptBackbone=llama-2-7b-chat-hf, Graph Integration=graph prompt tuning2025.06 | 1.5 | 44.44 | 21.01 | 32.95 | 39.97 | 20.42 | 28.01 | 1.01 | 2.5 | |
| LED + graph promptBackbone=led-large-16384, Graph Integration=graph prompt tuning2025.06 | 1.33 | 42.06 | 19.96 | 31.74 | 37.56 | 18.23 | 29.84 | 0.84 | 2.17 | |
| LED + textual graphBackbone=led-large-16384, Graph Integration=textual graph2025.06 | 1.29 | 41.84 | 19.46 | 31.18 | 37.3 | 18.23 | 29.77 | 0.83 | 2.12 | |
| LED + both (full model)Backbone=led-large-16384, Graph Integration=textual graph + graph prompt2025.06 | 1.26 | 42.96 | 20.66 | 32.74 | 38.56 | 19.09 | 28.14 | 0.71 | 1.97 |