Multi-turn response selection on Ubuntu Dialogue Corpus V2 (test)
0.859R10@1Uni-Encoder
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Uni-EncoderContext-Response Full Attention=true, Avoidance of Context Recomputation=true, Pre-trained Weights=BERT2021.06 | 0.859 | — | — | — | |
| Uni-EncoderProtocol=fine-tuned2021.06 | 0.859 | — | 0.938 | 0.99 | |
| SA-BERTspeaker_aware=true2020.04 | 0.83 | 0.963 | 0.919 | 0.985 | |
| SA-BERTProtocol=fine-tuned2021.06 | 0.83 | — | 0.919 | 0.985 | |
| Cross-EncoderContext-Response Full Attention=true, Avoidance of Context Recomputation=false, Pre-trained Weights=BERT2021.06 | 0.828 | — | — | — | |
| Poly-EncoderContext-Response Full Attention=false, Avoidance of Context Recomputation=true, Pre-trained Weights=BERT2021.06 | 0.809 | — | — | — | |
| Poly-Encoder 360Protocol=fine-tuned2021.06 | 0.809 | — | — | 0.981 | |
| Bi-EncoderContext-Response Full Attention=false, Avoidance of Context Recomputation=true, Pre-trained Weights=BERT2021.06 | 0.806 | — | — | — | |
| BERTspeaker_aware=false2020.04 | 0.781 | 0.95 | 0.89 | 0.98 | |
| BERTProtocol=fine-tuned2021.06 | 0.781 | — | 0.89 | 0.98 | |
| IMN2020.04 | 0.771 | 0.945 | 0.886 | 0.979 |