Table 6 Performance of the association prediction methods using character context-based embeddings learned from both primary-character and secondary-character tags
From: A multi-modal dataset and method for bone-level association prediction in oracle bone inscriptions
Model Type | Method | Evaluation Metric | ||||||
|---|---|---|---|---|---|---|---|---|
Sentence Embedding | Classifier | AUROC | AUPR | Accuracy | Precision | Recall | F1 score | |
Two-Stage | SIF | LR | 0.5446 | 0.1271 | 0.5212 | 0.1002 | 0.5341 | 0.1687 |
SVM | 0.5502 | 0.1504 | 0.6634 | 0.1193 | 0.4233 | 0.1861 | ||
XGBoost | 0.7793 | 0.3959 | 0.9176 | 0.7199 | 0.1540 | 0.2537 | ||
LightGBM | 0.7559 | 0.3490 | 0.9132 | 0.8070 | 0.0593 | 0.1104 | ||
MLP | 0.7222 | 0.2464 | 0.9095 | 0.5846 | 0.0171 | 0.0332 | ||
uSIF | LR | 0.5457 | 0.1373 | 0.9071 | 0.2927 | 0.0155 | 0.0294 | |
SVM | 0.5711 | 0.1203 | 0.8333 | 0.1496 | 0.1778 | 0.1625 | ||
XGBoost | 0.7738 | 0.4021 | 0.9191 | 0.7268 | 0.1765 | 0.2841 | ||
LightGBM | 0.7657 | 0.3763 | 0.9148 | 0.8356 | 0.0786 | 0.1437 | ||
MLP | 0.7451 | 0.2748 | 0.9095 | 0.6126 | 0.0160 | 0.0311 | ||
End-to-End | TextCNN | 0.9089 | 0.5847 | 0.9295 | 0.6497 | 0.5147 | 0.5692 | |
Transformer Encoder | 0.8143 | 0.3368 | 0.9120 | 0.5780 | 0.1272 | 0.2073 | ||
LSTM | 0.8987 | 0.5438 | 0.9210 | 0.5728 | 0.5222 | 0.5456 | ||
BiLSTM | 0.9132 | 0.5968 | 0.9294 | 0.6207 | 0.5758 | 0.5973 | ||
SGBSAP | 0.9535 | 0.7592 | 0.9481 | 0.7305 | 0.6812 | 0.7050 | ||
