Table 8 Ablation study on GoEmotions dataset.
From: A weighted difference loss approach for enhancing multi-label classification
Model | MF1 (%) \(\uparrow\) | mF1 (%) \(\uparrow\) | Precision (%) \(\uparrow\) | Recall (%) \(\uparrow\) |
|---|---|---|---|---|
SAN + WDL3 | 51.56 | 60.63 | 54.14 | 65.51 |
SAN + WDL2 | 52.13 | 60.63 | 58.18 | 53.32 |
SAN + WDL1 | 53.55 | 60.58 | 53.57 | 58.07 |
SAN + WDL0 | 53.08 | 59.96 | 51.14 | 60.16 |
WDL2 | 51.68 | 61.58 | 57.05 | 51.33 |
WDL1 | 52.18 | 60.98 | 52.30 | 55.12 |
WDL0 | 51.80 | 60.85 | 50.69 | 55.65 |
D2 | 48.37 | 59.54 | 48.00 | 53.64 |
D1 | 50.27 | 59.50 | 48.84 | 55.78 |
D0 (label + ratio) | 51.59 | 59.93 | 49.73 | 56.57 |
Bert | 44.81 | 59.22 | 56.46 | 39.84 |
