Table 7.
Evaluation results for the Hallmarks of Cancer task (HOC) text classification task
| Document classification | Sentence classification | |||||
|---|---|---|---|---|---|---|
| Model | Precision | Recall | F1 | Precision | Recall | F1 |
| Baseline (no retrofitting) | 77.8 | 51.7 | 62.1 | 56.8 | 30.7 | 39.9 |
| 22-classes retrofitted | 74.4 | 62.1 | 67.7* | 49.1 | 35.8 | 41.4* |
| 117-subclasses retrofitted | 74.8 | 62.5 | 68.1* | 48.6 | 35.2 | 40.8* |
The Baseline model is a skip-gram model without any retrofitting. All figures are micro-averages expressed as percentages (Bold denotes the best F1-score, * denotes statistically significant scores with respect to the baseline)