Feature Attribution on Rotten Tomatoes fine-tuned
-0.121LOGrad*Inp
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Grad*InpModel=RoBERTa2023.05 | -0.121 | 0.0617 | 0.363 | |
| DeepLiftModel=RoBERTa2023.05 | -0.163 | 0.0804 | 0.348 | |
| Grad*InpModel=DistilBERT2023.05 | -0.257 | 0.0681 | 0.315 | |
| DeepLiftModel=DistilBERT2023.05 | -0.332 | 0.101 | 0.26 | |
| GradientShapModel=RoBERTa2023.05 | -0.389 | 0.194 | 0.299 | |
| DIGModel=RoBERTa2023.05 | -0.426 | 0.223 | 0.286 | |
| Grad*InpModel=BERT2023.05 | -0.438 | 0.143 | 0.438 | |
| GradientShapModel=DistilBERT2023.05 | -0.452 | 0.237 | 0.212 | |
| DeepLiftModel=BERT2023.05 | -0.452 | 0.123 | 0.45 | |
| DIGModel=DistilBERT2023.05 | -0.487 | 0.273 | 0.181 | |
| SIGModel=DistilBERT2023.05 | -0.533 | 0.331 | 0.134 | |
| IGModel=DistilBERT2023.05 | -0.54 | 0.341 | 0.163 | |
| GradientShapModel=BERT2023.05 | -0.715 | 0.204 | 0.438 | |
| IGModel=RoBERTa2023.05 | -0.787 | 0.354 | 0.242 | |
| SIGModel=RoBERTa2023.05 | -0.869 | 0.361 | 0.251 | |
| DIGModel=BERT2023.05 | -1.05 | 0.293 | 0.414 | |
| IGModel=BERT2023.05 | -1.19 | 0.307 | 0.41 | |
| SIGModel=BERT2023.05 | -1.52 | 0.39 | 0.349 |