Table 8 Performance comparison of downstream layers
From: BertSRC: transformer-based semantic relation classification
Layer architecture | Accuracy | Precision | Recall | F1-score |
|---|---|---|---|---|
CLS token layer | *0.858 **(0.016) | 0.823 (0.026) | 0.835 (0.015) | 0.827 (0.017) |
Two-token layer | 0.867 (0.014) | 0.840 (0.018) | 0.847 (0.011) | 0.842 (0.013) |
Three-token layer | 0.861 (0.013) | 0.835 (0.027) | 0.842 (0.009) | 0.836 (0.016) |