Table 2.
Performance comparison
| Corpus | HPRD50 | IEPA | LLL | PICADb | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| F | P | R | F | P | R | F | P | R | F | P | R | |
| Bui et al. [24] | 71.7 | 62.2 | 84.7 | 73.4 | 62.9 | 88.1 | 83.6 | 81.9 | 85.4 | – | – | – |
| Miwa et al. [49] | 70.9 | 68.5 | 76.1 | 71.7 | 67.5 | 78.6 | 80.1 | 77.6 | 86.0 | – | – | – |
| Chang et al. [48] | 71.5 | 63.8 | 81.2 | 71.4 | 62.5 | 83.3 | 80.6 | 73.2 | 89.6 | – | – | – |
| Murugesan et al. [73] | 80.0 | 76.3 | 84.2 | 80.2 | 75.9 | 85.2 | 89.2 | 87.3 | 91.2 | – | – | – |
| aZhao et al. [81] | 71.3 | 58.7 | 92.4 | 74.2 | 67.0 | 84.0 | 82.0 | 75.8 | 91.8 | – | – | – |
| GRGT | 64.0 | 86.5 | 50.8 | 74.9 | 91.0 | 63.6 | 83.6 | 91.2 | 77.1 | 70.0 | 78.2 | 63.4 |
Performance comparison of our method (GRGT) with top-performing methods on four benchmark datasets. F F1-score, P precision, R recall. The measurement is out of 100. adeep learning method. bValues are not available because of the unavailability of executable program or source code