Controllable Text Generation on IMDb (test)
10.89O-PPLPT(select)+PL
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PT(select)+PLMethod Category=Self-Training with PLM, Backbone PLM=GPT2, Pseudo Labeler=BERT-base2022.12 | 10.89 | 33.89 | 0.8875 | 0.2717 | 0.7141 | — | — | |
| GPT2 + PT(select) + PLBase Model=GPT2, Strategy=Self-Training, Components=Pseudo-Text Selection, Pseudo-Labels2022.12 | 10.89 | 33.89 | 88.75 | 27.17 | 71.41 | 88.32 | 96.24 | |
| PT(noise)+PLMethod Category=Self-Training with PLM, Backbone PLM=GPT2, Pseudo Labeler=BERT-base2022.12 | 11.26 | 33.85 | 0.8847 | 0.2726 | 0.709 | — | — | |
| GPT2 + PT(noise) + PLBase Model=GPT2, Strategy=Self-Training, Components=Pseudo-Text with Noise, Pseudo-Labels2022.12 | 11.26 | 33.85 | 88.47 | 27.26 | 70.9 | 87.6 | 95.59 | |
| PT(noise)Method Category=Self-Training with PLM, Backbone PLM=GPT22022.12 | 11.91 | 44.31 | 0.7746 | 0.254 | 0.7219 | — | — | |
| GPT2 + PT + noiseBase Model=GPT2, Strategy=Self-Training, Components=Pseudo-Text, Noise2022.12 | 11.91 | 44.31 | 77.46 | 25.4 | 72.19 | 74.95 | 85.02 | |
| Ctr-PFMethod Category=Lightweight method2022.12 | 13.01 | 37.12 | 0.7733 | 0.2963 | 0.6483 | — | — | |
| Ctr-PFCategory=Lightweight method2022.12 | 13.01 | 37.12 | 77.33 | 29.63 | 64.83 | 71 | 86.51 | |
| PFMethod Category=Lightweight method2022.12 | 13.02 | 37.09 | 0.7505 | 0.2948 | 0.651 | — | — | |
| PFCategory=Lightweight method2022.12 | 13.02 | 37.09 | 75.05 | 29.48 | 65.1 | 67.55 | 81.84 | |
| GPT2(raw)Method Category=Zero-shot PLM2022.12 | 13.2 | 38.39 | 0.685 | 0.3591 | 0.5879 | — | — | |
| GPT2(raw)Mode=Raw2022.12 | 13.2 | 38.39 | 68.5 | 35.91 | 58.79 | 55.9 | 61.37 | |
| PTMethod Category=Self-Training with PLM, Backbone PLM=GPT22022.12 | 14.62 | 68.04 | 0.7957 | 0.3058 | 0.6522 | — | — | |
| GPT2 + PTBase Model=GPT2, Strategy=Self-Training, Components=Pseudo-Text2022.12 | 14.62 | 68.04 | 79.57 | 30.58 | 65.22 | 76.1 | 87.92 | |
| GPT2Method Category=Finetuned PLM2022.12 | 16.4 | 44.02 | 0.8044 | 0.2634 | 0.71 | — | — | |
| GPT2Category=Finetune LM2022.12 | 16.4 | 44.02 | 80.44 | 26.34 | 71 | 77.55 | 88.35 | |
| UniLM + PT(select) + PLBase Model=UniLM, Strategy=Self-Training, Components=Pseudo-Text Selection, Pseudo-Labels2022.12 | 18.4 | 33.56 | 90.06 | 31.27 | 67.61 | 90.08 | 96.66 | |
| UniLM + PT(noise) + PLBase Model=UniLM, Strategy=Self-Training, Components=Pseudo-Text with Noise, Pseudo-Labels2022.12 | 18.92 | 33.53 | 89.73 | 30.94 | 66.84 | 89.95 | 96.38 | |
| DuNSTMethod Category=Our method2022.12 | 21.67 | 42.82 | 0.9305 | 0.3179 | 0.658 | — | — | |
| DuNST2022.12 | 21.67 | 42.82 | 93.05 | 31.79 | 65.8 | 92.9 | 98.02 | |
| Ground TruthMethod Category=Reference2022.12 | 25.14 | — | 0.962 | 0.4827 | 0.4334 | — | — | |
| UniLMMethod Category=Finetuned PLM2022.12 | 25.2 | 54.33 | 0.7535 | 0.3105 | 0.6697 | — | — | |
| UniLMCategory=Finetune LM2022.12 | 25.2 | 54.33 | 75.35 | 31.05 | 66.97 | 76.45 | 85.18 | |
| T5Method Category=Finetuned PLM2022.12 | 25.69 | 34.97 | 0.8377 | 0.3003 | 0.6957 | — | — | |
| T5Category=Finetune LM2022.12 | 25.69 | 34.97 | 83.77 | 30.03 | 69.57 | 82.8 | 90.5 | |
| UniLM + PTBase Model=UniLM, Strategy=Self-Training, Components=Pseudo-Text2022.12 | 26.62 | 58.37 | 70.27 | 31.17 | 66.69 | 72.2 | 80.37 | |
| UniLM + PT + noiseBase Model=UniLM, Strategy=Self-Training, Components=Pseudo-Text, Noise2022.12 | 30.28 | 62.07 | 75.78 | 31.68 | 65.18 | 77.75 | 85.35 |