Table 4 Performance of the attention-based joint learning model with Siamese network extracted embeddings on ATIS and SNIPS datasets across different triplet-loss margins (mean ± standard deviation over five runs).
Models | Margin | ATIS | SNIPS | ||
|---|---|---|---|---|---|
Accuracy (%) | F1-score (%) | Accuracy (%) | F1-score (%) | ||
SBJLIS_1 | 0.1 | 98.87 ± 0.02 | 98.60 ± 0.04 | 99.23 ± 0.01 | 98.43 ± 0.03 |
SBJLIS_2 | 0.2 | 97.74 ± 0.03 | 97.92 ± 0.05 | 99.00 ± 0.02 | 98.02 ± 0.03 |
SBJLIS_3 | 0.3 | 96.74 ± 0.04 | 98.05 ± 0.04 | 98.95 ± 0.02 | 98.30 ± 0.03 |
SBJLIS_4 | 0.4 | 97.11 ± 0.03 | 98.13 ± 0.05 | 98.93 ± 0.03 | 97.86 ± 0.04 |
SBJLIS_5 | 0.5 | 96.86 ± 0.04 | 98.43 ± 0.05 | 98.89 ± 0.03 | 97.83 ± 0.04 |
SBJLIS_6 | 0.6 | 96.49 ± 0.03 | 97.66 ± 0.05 | 98.58 ± 0.04 | 97.82 ± 0.03 |
SBJLIS_7 | 0.7 | 97.11 ± 0.02 | 98.06 ± 0.04 | 98.50 ± 0.03 | 97.81 ± 0.04 |
SBJLIS_8 | 0.8 | 96.74 ± 0.04 | 98.30 ± 0.05 | 98.20 ± 0.04 | 97.29 ± 0.04 |
SBJLIS_9 | 0.9 | 95.36 ± 0.05 | 97.73 ± 0.06 | 98.07 ± 0.04 | 97.86 ± 0.04 |
SBJLIS_10 | 1.0 | 96.49 ± 0.05 | 98.21 ± 0.05 | 97.63 ± 0.04 | 97.85 ± 0.05 |
