Response Generation on ChattyChef (test)
5.4BLEUChatGPT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ChatGPTsubset=evaluated on human evaluation subset (10 conversations)2023.05 | 5.4 | 53 | 64.9 | 12.5 | 45.3 | |
| GPT-J+ctrgrounded knowledge selection=Center approach2023.05 | 4.7 | 45.9 | 11.7 | 9.3 | 36.6 | |
| GPT-J+cutgrounded knowledge selection=Cut approach2023.05 | 4.3 | 45.2 | 10.9 | 9.9 | 38.7 | |
| GPT-J+ctr+intstate-aware model=true, grounded knowledge selection=Center approach, incorporates=User Intent2023.05 | 4.2 | 45.1 | 10.3 | 10.8 | 39.3 | |
| GPT-J2023.05 | 4.1 | 44.7 | 11.1 | 9.9 | 37.9 | |
| GPT-J+intincorporates=User Intent information2023.05 | 3.9 | 45 | 10 | 10.4 | 38.5 |