Skip to Main Content
Table 7: 
Accuracy in classifying Pos vs Rand Neg and Pos vs Adv Neg responses for various model variants trained/finetuned on DailyDialog++.
ModelTraining/Finetuning DataPos vs Rand NegPos vs Adv Neg
BERT regressor Rand neg 73.40 67.57 
Adv neg 69.89 75.92 
Rand + Adv neg 72.77 74.55 
 
BERT+DNN Rand neg 74.67 60.14 
Adv neg 60.49 87.67 
Rand + Adv neg 73.87 86.61 
 
RUBER (Pretrained) Rand neg 78.18 64.96 
Adv neg 70.82 76.50 
Rand + Adv neg 75.11 83.88 
 
RUBER-Large (Pretrained) Rand neg 82.35 68.94 
Adv neg 63.99 90.49 
Rand + Adv neg 79.91 86.54 
 
DEB (Pretrained) Rand neg 88.29 66.75 
Adv neg 86.24 82.04 
Rand + Adv neg 88.67 92.65 
ModelTraining/Finetuning DataPos vs Rand NegPos vs Adv Neg
BERT regressor Rand neg 73.40 67.57 
Adv neg 69.89 75.92 
Rand + Adv neg 72.77 74.55 
 
BERT+DNN Rand neg 74.67 60.14 
Adv neg 60.49 87.67 
Rand + Adv neg 73.87 86.61 
 
RUBER (Pretrained) Rand neg 78.18 64.96 
Adv neg 70.82 76.50 
Rand + Adv neg 75.11 83.88 
 
RUBER-Large (Pretrained) Rand neg 82.35 68.94 
Adv neg 63.99 90.49 
Rand + Adv neg 79.91 86.54 
 
DEB (Pretrained) Rand neg 88.29 66.75 
Adv neg 86.24 82.04 
Rand + Adv neg 88.67 92.65 
Close Modal

or Create an Account

Close Modal
Close Modal