Table 16 Wilcoxon signed-rank test results for the Hypert and BERT models with 10 iterations
From: Hypert: hypernymy-aware BERT with Hearst pattern exploitation for hypernym discovery
| Â | Subtask | Evaluation measures | BERT model | |
|---|---|---|---|---|
Hypert model versus (p-value) | 1A | MRR | Win | (9.76e–3) |
MAP | Win | (3.90e–3) | ||
P@1 | Win | (9.76e–3) | ||
P@3 | Win | (2.73e–2) | ||
P@5 | Win | (3.90e–3) | ||
P@15 | Win | (3.71e–2) | ||
2A | MRR | Win | (4.88e–2) | |
MAP | Win | (2.73e–2) | ||
P@1 | Win | (4.88e–2) | ||
P@3 |  | (1.30e–1) | ||
P@5 |  | (7.67e–1) | ||
P@15 | Win | (4.88e–2) | ||
2B | MRR | Win | (4.88e–2) | |
MAP | Win | (9.76e–3) | ||
P@1 |  | (6.44e–2) | ||
P@3 | Win | (3.9e–3) | ||
P@5 | Win | (1.36e–2) | ||
P@15 | Win | (3.9e–3) | ||