Table 8.
Performance comparison (maximum F score) with existing models on the DDI corpus
| Methods | Team | F | R | P |
|---|---|---|---|---|
| SVM-based | Uturku [51] | 59.4 | 49.9 | 73.2 |
| FBK-irst [52] | 65.1 | 65.6 | 64.6 | |
| RAIHANI [53] | 71.1 | 68.7 | 73.7 | |
| DL-based | Liu-CNN [9] | 69.8 | 64.7 | 75.7 |
| MCCNN [54] | 70.2 | 65.3 | 76.0 | |
| TEES-CNN [55] | 73.5 | 67.6 | 80.5 | |
| Joint AB-LSTM [13] | 71.5 | 65.0 | 74.5 | |
| Char-RNNs [42] | 72.1 | 65.9 | 80.0 | |
| Hierarchy RNN [14] | 72.9 | 71.8 | 74.1 | |
| Recursive NN [56] | 73.5 | 69.6 | 77.8 | |
| Attn-BLSTM [30] | 77.3 | 76.8 | 78.4 | |
| Our model | 80.3 | 83.0 | 77.6 |
The listed results derive from the corresponding papers. F: F score (micro); R: recall; P: precision. The highest scores are highlighted in bold