Sentiment Analysis on NSMC (Robustness Evaluation)
89.4Accuracy (Clean)BERT
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| BERTTokenization=MorSubword2026.04 | 89.4 | 87.57 | 85.72 | 85.3 | 79.94 | 87.66 | 85.63 | 85.29 | 79.93 | |
| BERTTokenization=Subword2026.04 | 88.91 | 87.41 | 85.86 | 84.84 | 79.56 | 87.31 | 85.57 | 84.9 | 79.73 | |
| BERTTokenization=Character2026.04 | 88.89 | 87.69 | 86.2 | 84.72 | 79.44 | 87.62 | 86.24 | 84.95 | 79.89 | |
| KOMBOTokenization=Jamo2026.04 | 88.7 | 87.57 | 86.33 | 84.49 | 79.15 | 87.8 | 86.7 | 84.64 | 79.44 | |
| BERTTokenization=Jamo2026.04 | 87.78 | 87.22 | 86.58 | 84.04 | 79.17 | 87.22 | 86.68 | 84.08 | 79.45 |