Interactive Human Evaluation on 100 negative situations (test)
67.5Fluency Win RateSUPPORTER
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SUPPORTERComparison Opponent=BlenderBot-Joint2023.07 | 67.5 | 23.7 | 8.8 | 55.2 | 40.7 | 4.1 | 53.8 | 31.8 | 14.4 | 59.2 | 34.1 | 6.7 | 56.5 | 30.4 | 13.1 | |
| SUPPORTERComparison Opponent=Bart-Joint2023.07 | 66.5 | 26.5 | 7 | 56.7 | 38.8 | 4.5 | 45.4 | 43.8 | 10.8 | 51.4 | 37.6 | 11 | 48.6 | 37.1 | 14.3 | |
| SUPPORTERComparison Opponent=w/o EmoExperts2023.07 | 44.5 | 40 | 15.5 | 48.6 | 36.8 | 14.6 | 53.7 | 35.7 | 10.6 | 54.5 | 33.4 | 12.1 | 50 | 34.3 | 15.7 | |
| SUPPORTERComparison Opponent=w/o ExpertPolicy2023.07 | 42.9 | 37.5 | 19.6 | 38.5 | 35.9 | 25.6 | 55.1 | 32.4 | 12.5 | 51.4 | 34.3 | 14.3 | 49.6 | 32.1 | 18.3 |