Relation extraction results on the CoNLL04 test partition. We used the pre-trained SpanBERT-large. Our full model trains on the entire training partition using the SSL method discussed in Section 3.2.1. The “burn-in only” setting trains just on the training subset that has annotations from rules.
| Approach . | Precision . | Recall . | F1 . |
|---|---|---|---|
| Baselines | |||
| Rules | 81.6 | 16.82 | 27.90 |
| SpanBERT (Joshi et al. 2020) | 81.30 ± 4.89 | 71.01 ± 5.11 | 75.78 ± 4.79 |
| Unsupervised Rationale | 83.91 ± 2.88 | 74.88 ± 1.44 | 79.11 ± 1.01 |
| Our Approach | |||
| Burn-in Only | 62.71 ± 2.27 | 53.32 ± 0.95 | 57.63 ± 1.39 |
| Full Model | 83.01 ± 2.16 | 76.30 ± 3.08 | 79.46 ± 0.92 |
| Approach . | Precision . | Recall . | F1 . |
|---|---|---|---|
| Baselines | |||
| Rules | 81.6 | 16.82 | 27.90 |
| SpanBERT (Joshi et al. 2020) | 81.30 ± 4.89 | 71.01 ± 5.11 | 75.78 ± 4.79 |
| Unsupervised Rationale | 83.91 ± 2.88 | 74.88 ± 1.44 | 79.11 ± 1.01 |
| Our Approach | |||
| Burn-in Only | 62.71 ± 2.27 | 53.32 ± 0.95 | 57.63 ± 1.39 |
| Full Model | 83.01 ± 2.16 | 76.30 ± 3.08 | 79.46 ± 0.92 |