Table 2.
Performance in terms of precision (P), recall (R), and F1, with standard deviation (SD) for weakly supervised relation extraction compared with baselines
| MODEL | Pain-Anatomy (n = 236) | Implant-Complication (n = 276) | ||||
|---|---|---|---|---|---|---|
| P (SD) | R (SD) | F1 (SD) | P (SD) | R (SD) | F1 (SD) | |
| Soft Majority Vote (SMV) | 81.4 (2.8) | 64.8 (3.0) | 72.1 (2.3) | 81.6 (3.6) | 31.7 (2.7) | 45.6 (3.1) |
| Fully Supervised (FS) | 72.5 (2.9) | 78.3 (2.6) | 75.3 (2.1) | 50.8 (3.1) | 47.1 (3.1) | 48.8 (2.7) |
| Weakly Supervised (WS) | 80.2 (2.6) | 82.5 (2.4) | 81.3 (1.9) | 82.6 (2.6) | 61.1 (2.9) | 70.2 (2.3) |
| Improvement over SMV | −1.5% | +27.3% | +12.8% | +1.2% | +92.7% | +53.9% |
Bold highlights show highest value for a given metric