Sentiment Analysis Data Stream (imdb, eng, zho, ara, spa, jpn, rus) (test)
89.95IMDB AccCTTA-T
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CTTA-TBackbone=Bert-base, Adaptation Mode=Continual Test-Time Adaptation, Optimizer=AdamW, Batch Size=16, Learning Rate=1e-52025.12 | 89.95 | 92.12 | 59.45 | 51.06 | 78.38 | 59.26 | 65.02 | 73.42 | |
| SoTTABackbone=Bert-base, Adaptation Mode=Continual Test-Time Adaptation2025.12 | 87.69 | 92.03 | 53.96 | 48.84 | 74.52 | 35.62 | 57.46 | 67.73 | |
| baseBackbone=Bert-base, Adaptation Mode=None (Zero-shot)2025.12 | 87.64 | 92.01 | 58.71 | 50.63 | 77.05 | 44.08 | 64.81 | 70.94 | |
| CoTTABackbone=Bert-base, Adaptation Mode=Continual Test-Time Adaptation2025.12 | 87.63 | 90.26 | 54.31 | 48.96 | 72.31 | 34.62 | 58.02 | 67.17 | |
| RAccBackbone=Bert-base, Adaptation Mode=Continual Test-Time Adaptation2025.12 | 87.02 | 91.25 | 52.98 | 47.63 | 72.13 | 34.91 | 55.12 | 66.61 | |
| SARBackbone=Bert-base, Adaptation Mode=Continual Test-Time Adaptation2025.12 | 86.75 | 91.59 | 54.58 | 49.45 | 76.95 | 39.86 | 61.57 | 69.12 |