Learning curve of our approach on TACRED based on amount of rules used. In each experiment, we use up to top k rules per relation type; the number in parentheses is the actual average number of rules per type.
| Num of Rules . | Precision . | Recall . | F1 . |
|---|---|---|---|
| Relation Classification | |||
| Up to top 1 (0.98 rules/relation) | 72.48 | 66.23 | 69.21 |
| Up to top 5 (3.56 rules/relation) | 72.97 | 69.02 | 70.94 |
| Up to top 10 (5.02 rules/relation) | 69.30 | 71.64 | 70.45 |
| All rules (7.27 rules/relation) | 71.15 | 71.13 | 71.14 |
| Explainability Classification | |||
| Up to top 1 (0.98 rules/relation) | 74.62 | 85.35 | 75.02 |
| Up to top 5 (3.56 rules/relation) | 92.19 | 94.06 | 91.28 |
| Up to top 10 (5.02 rules/relation) | 91.06 | 95.62 | 91.22 |
| All rules (7.27 rules/relation) | 95.63 | 97.92 | 95.76 |
| Num of Rules . | Precision . | Recall . | F1 . |
|---|---|---|---|
| Relation Classification | |||
| Up to top 1 (0.98 rules/relation) | 72.48 | 66.23 | 69.21 |
| Up to top 5 (3.56 rules/relation) | 72.97 | 69.02 | 70.94 |
| Up to top 10 (5.02 rules/relation) | 69.30 | 71.64 | 70.45 |
| All rules (7.27 rules/relation) | 71.15 | 71.13 | 71.14 |
| Explainability Classification | |||
| Up to top 1 (0.98 rules/relation) | 74.62 | 85.35 | 75.02 |
| Up to top 5 (3.56 rules/relation) | 92.19 | 94.06 | 91.28 |
| Up to top 10 (5.02 rules/relation) | 91.06 | 95.62 | 91.22 |
| All rules (7.27 rules/relation) | 95.63 | 97.92 | 95.76 |