Relation Extraction on DNRTI-AUG-STIX2 (test)
98PrecisionSecureBERT+-BiGRU-CRF
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SecureBERT+-BiGRU-CRFbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, hidden layer size=100 × 2, embedding size=768, number of epochs=4, optimizer=AdamW2026.05 | 98 | 97 | 97 | |
| SecureBERT+-BiGRU-TDDbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, hidden layer size=100 × 2, embedding size=768, number of epochs=4, optimizer=AdamW2026.05 | 98 | 98 | 98 | |
| SecureBERT-BiGRU-CRFbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, hidden layer size=100 × 2, embedding size=768, number of epochs=4, optimizer=AdamW2026.05 | 97 | 97 | 97 | |
| RobERTa-BiGRU-TDD2026.05 | 95 | 94 | 94 | |
| RobERTa-BiGRU-CRF2026.05 | 94 | 94 | 94 | |
| Glove-BiLSTM-TDD+Attbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, hidden layer size=100 × 2, embedding size=300, number of epochs=50, optimizer=AdamW2026.05 | 91 | 91 | 91 | |
| BERT-BiLSTM-TDD+Attbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, hidden layer size=100 × 2, embedding size=768, number of epochs=4, optimizer=AdamW2026.05 | 90 | 89 | 89 | |
| Glove-CNN-TDD+Attbatch size=16, dropout=0.3, epsilon=1e-8, initial learning rate=5e-5, embedding size=300, number of epochs=100, optimizer=AdamW, window=3, pos_dim=5, pos_dist=502026.05 | 84 | 84 | 84 |