Table 6 Model performance with different negative sampling methods
From: B-LBConA: a medical entity disambiguation model based on Bio-LinkBERT and context-aware mechanism
Method | NCBI | ADR | ShARe/CLEF | |||
|---|---|---|---|---|---|---|
test | refined test | test | refined test | test | refined test | |
Random Negative Sampling [40] | 78.26 | 52.98 | 75.84 | 50.49 | 80.96 | 60.74 |
Popularity-biased Negative Sampling [41] | 79.32 | 50.30 | 80.21 | 55.21 | 78.39 | 58.23 |
Adversarial Negative Sampling [42] | 85.66 | 70.76 | 85.43 | 67.38 | 86.52 | 69.79 |
Dynamically Negative Sampling [43] | 90.47 | 81.13 | 92.03 | 80.67 | 91.65 | 77.18 |
Our sampling method | 93.57 | 74.38 | 94.72 | 79.89 | 94.23 | 80.68 |