Unsupervised Abstractive Summarization on CNN/DM (test)
40.13ROUGE-1FAR-RW*
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FAR-RW*source=Zhang et al., 2022, SOTA=true2022.12 | 40.13 | 17 | 36.34 | — | |
| Self-training (3rd round) + SummScoreDecoding=Beam search, Beams=20, Iteration=3, Re-ranking=SummScore2022.12 | 39.76 | 16.79 | 35.85 | 87.18 | |
| Self-training (2nd round) + SummScoreDecoding=Beam search, Beams=20, Iteration=2, Re-ranking=SummScore2022.12 | 39.49 | 16.69 | 35.61 | 87.07 | |
| Self-training (1st round) + SummScoreDecoding=Beam search, Beams=20, Iteration=1, Re-ranking=SummScore2022.12 | 38.75 | 16.11 | 34.78 | 86.88 | |
| TEDsource=Yang et al., 20202022.12 | 38.73 | 16.84 | 35.4 | — | |
| Self-training (3rd round)Decoding=Beam search, Beams=20, Iteration=32022.12 | 38.47 | 15.95 | 34.48 | 87 | |
| Self-training (2nd round)Decoding=Beam search, Beams=20, Iteration=22022.12 | 38.17 | 15.77 | 34.25 | 86.87 | |
| Summary Loop 45source=Laban et al., 20202022.12 | 37.7 | 14.8 | 34.7 | — | |
| PEGASUS (ours) + SummScoreDecoding=Beam search, Beams=20, Re-ranking=SummScore2022.12 | 36.92 | 15.03 | 33.19 | 86.54 | |
| Self-training (1st round)Decoding=Beam search, Beams=20, Iteration=12022.12 | 36.68 | 14.52 | 32.72 | 86.49 | |
| PEGASUS (ours)Decoding=Beam search, Beams=202022.12 | 35.47 | 13.89 | 31.61 | 86.29 | |
| PEGASUSsource=Zhang et al., 20202022.12 | 32.9 | 13.28 | 29.38 | — |