Dialogue Evaluation on DailyDialog (eval)
0.579Spearman CorrelationMDD-S
Evaluation results
| Method | Links | |
|---|---|---|
| MDD-STraining Dataset=MDD-Data, Size=~600K, Pretrained Model=RoBERTa, Objective=CrossEntropy / MLM / KL, External Knowledge=None, Single=Yes, Description=full student model optimized with all three losses2021.12 | 0.579 | |
| MDD-CMObjective=L_MLM and L_CE, Description=student model optimized with both L_MLM and L_CE2021.12 | 0.546 | |
| GRADETraining Dataset=DailyDialog, Size=~178K, Pretrained Model=BERT, Objective=Triplet, External Knowledge=ConceptNet, Single=Yes2021.12 | 0.533 | |
| MDD-TTraining Dataset=DailyDialog++, Size=~139K, Pretrained Model=RoBERTa, Objective=CrossEntropy, External Knowledge=None, Single=Yes, Role=Teacher model2021.12 | 0.501 | |
| DEBTraining Dataset=DailyDialog++, Size=~139K, Pretrained Model=BERT, Objective=CrossEntropy, External Knowledge=Reddit Conversations, Single=Yes2021.12 | 0.486 | |
| MDD-CObjective=L_CE, Description=student model optimized with only L_CE2021.12 | 0.482 | |
| D-scoreTraining Dataset=PersonaChat / Twitter / DailyDialog, Size=~1.31M/~1.16M, Pretrained Model=RoBERTa, Objective=MLM / CrossEntropy, External Knowledge=None, Single=No2021.12 | 0.426 | |
| USL-HTraining Dataset=DailyDialog, Size=~138K, Pretrained Model=BERT, Objective=MLM / CrossEntropy, External Knowledge=None, Single=No2021.12 | 0.391 | |
| USRTraining Dataset=TopicalChat / PersonaChat, Size=Unknown, Pretrained Model=RoBERTa, Objective=MLM / CrossEntropy, External Knowledge=Persona Profiles / Wikipedia, Single=No2021.12 | 0.367 | |
| uBERT-RTraining Dataset=DailyDialog / PersonaChat, Size=Unknown, Pretrained Model=BERT, Objective=Triplet, External Knowledge=None, Single=Yes2021.12 | 0.285 |