Table 3 Performance comparison with other state-of-art methods on DDIExtraction 2013 dataset
| Â | Methods | Precision (%) | Recall (%) | F1-score (%) |
|---|---|---|---|---|
CNN-based | SCNN [12] | 69.1 | 65.1 | 67.0 |
DCNN [11] | 77.2 | 64.4 | 70.2 | |
MCCNN [10] | 76.0 | 65.3 | 70.2 | |
LSTM-based | DLSTM [18] | 72.5 | 71.5 | 72.0 |
ASDP-LSTM [20] | 74.1 | 71.8 | 72.9 | |
Tree-LSTM [37] | 77.8 | 69.6 | 73.5 | |
ATT-BLSTM [19] | 78.4 | 76.2 | 77.3 | |
Pretraining-based | BERT [29] | 77.90 | 77.43 | 77.66 |
BioBERT [31] | 81.1 | 75.3 | 78.1 | |
CharacterBERT [32] | 79.18 | 80.38 | 81.70 | |
ChemicalBERT + AGGCN [38] | 83.96 | 81.82 | 82.88 | |
DESC_MOL [39] | 84.69 | 82.53 | 83.60 | |
IMSE (ours) | 85.63 | 85.17 | 85.16 |