Knowledge-grounded Dialog Generation on Wizard of Wikipedia (WoW)
97.7Win RatePOKI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| POKIComparison Baseline=POKI w/o Entailment, Criteria=Engaging2022.03 | 97.7 | 0.8 | 0.7 | |
| POKIComparison Baseline=RAG, Criteria=Interesting2022.03 | 96.3 | 3.5 | 0.74 | |
| POKIComparison Baseline=RAG, Criteria=Coherent2022.03 | 95.4 | 4.5 | 0.78 | |
| POKIComparison Baseline=POKI w/o Entailment, Criteria=Coherent2022.03 | 94.3 | 3.4 | 0.68 | |
| POKIComparison Baseline=POKI w/o Kw Fidelity, Criteria=Interesting2022.03 | 93.5 | 4.5 | 0.71 | |
| POKIComparison Baseline=BART, Criteria=Humanlike2022.03 | 92.4 | 6.5 | 0.66 | |
| POKIComparison Baseline=RAG, Criteria=Humanlike2022.03 | 91.4 | 7.1 | 0.68 | |
| POKIComparison Baseline=RAG, Criteria=Engaging2022.03 | 89.3 | 7.7 | 0.72 | |
| POKIComparison Baseline=BART, Criteria=Coherent2022.03 | 88.5 | 9.6 | 0.72 | |
| POKIComparison Baseline=BART, Criteria=Engaging2022.03 | 87.8 | 8.3 | 0.71 | |
| POKIComparison Baseline=POKI w/o Entailment, Criteria=Humanlike2022.03 | 84.5 | 10.5 | 0.67 | |
| POKIComparison Baseline=POKI w/o Kw Fidelity, Criteria=Coherent2022.03 | 83.6 | 10.7 | 0.65 | |
| POKIComparison Baseline=BART, Criteria=Interesting2022.03 | 83.3 | 9.9 | 0.75 | |
| POKIComparison Baseline=POKI w/o Kw Fidelity, Criteria=Humanlike2022.03 | 81.8 | 13.5 | 0.71 | |
| POKIComparison Baseline=POKI w/o Entailment, Criteria=Interesting2022.03 | 79.8 | 17.2 | 0.7 | |
| POKIComparison Baseline=POKI w/o Kw Fidelity, Criteria=Engaging2022.03 | 71.5 | 25.4 | 0.69 | |
| POKIComparison Baseline=Gold, Criteria=Humanlike2022.03 | 42.3 | 41.9 | 0.68 | |
| POKIComparison Baseline=Gold, Criteria=Interesting2022.03 | 35.9 | 37.8 | 0.76 | |
| POKIComparison Baseline=Gold, Criteria=Engaging2022.03 | 25.4 | 26.7 | 0.73 | |
| POKIComparison Baseline=Gold, Criteria=Coherent2022.03 | 23.8 | 25.3 | 0.73 |