Sentiment Classification on Yelp
97.27AccuracyROBERTa-CL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ROBERTa-CLTraining supervision type=clean labels, Method categorization=Fully-supervised2020.10 | 97.27 | — | |
| MPD2024.06 | 96.4 | 0.1 | |
| COSINEMethod categorization=Framework2020.10 | 95.97 | — | |
| DP2O2024.06 | 94.3 | 0.4 | |
| TEMPERA2024.06 | 92.6 | 1.7 | |
| MixupMethod categorization=Baseline2020.10 | 92.05 | — | |
| USTMethod categorization=Baseline2020.10 | 90.53 | — | |
| SMARTMethod categorization=Baseline2020.10 | 88.58 | — | |
| DenoiseMethod categorization=Baseline2020.10 | 87.53 | — | |
| FreeLBMethod categorization=Baseline2020.10 | 85.68 | — | |
| InitMethod categorization=Framework2020.10 | 81.76 | — | |
| Self-ensembleMethod categorization=Baseline2020.10 | 80.08 | — | |
| WeSTClassMethod categorization=Baseline2020.10 | 76.86 | — | |
| ImplyLossMethod categorization=Baseline2020.10 | 76.29 | — | |
| ROBERTa-WL+Training supervision type=weak labels, Method categorization=Baseline2020.10 | 74.89 | — | |
| SnorkelMethod categorization=Baseline2020.10 | 69.21 | — | |
| ExMatchMethod categorization=Baseline2020.10 | 68.68 | — | |
| BiLSTM generalized poolingarchitecture=vector-based multi-head attention2018.06 | 66.55 | — | |
| BiLSTM mean poolingimplementation=Baseline implementation in this paper2018.06 | 65.3 | — | |
| BiLSTM max poolingimplementation=Baseline implementation in this paper2018.06 | 65 | — | |
| BiLSTM last poolingimplementation=Baseline implementation in this paper2018.06 | 64.95 | — | |
| BiLSTM self-attention (Lin et al., 2017)source=Lin et al., 20172018.06 | 64.21 | — | |
| CNN max pooling (Lin et al., 2017)source=Lin et al., 20172018.06 | 62.05 | — | |
| BiLSTM max pooling (Lin et al., 2017)source=Lin et al., 20172018.06 | 61.99 | — |