Skip to main content

Table 9 Comparison of for P@3, P@5, and P@15 results by number of pretraining steps

From: Hypert: hypernymy-aware BERT with Hearst pattern exploitation for hypernym discovery

Subtask

Steps

Evaluation measures

P@3

P@5

P@15

1A

English

0k

20.90 ± 1.00

20.68 ± 0.79

26.63 ± 0.75

1k

21.56 ± 1.38

21.27 ± 1.25

27.52 ± 1.35

2k

21.01 ± 1.05

20.83 ± 0.84

26.79 ± 1.16

3k

21.25 ± 0.99

21.04 ± 0.93

26.95 ± 1.11

4k

21.19 ± 0.94

20.88 ± 0.87

26.89 ± 0.95

5k

21.81 ± 1.25

21.61 ± 0.87

27.96 ± 0.57

6k

\(\underline{21.93 \pm 1.63}\)

21.58 ± 1.18

27.52 ± 0.96

7k

21.48 ± 1.16

21.12 ± 1.11

27.12 ± 1.42

8k

21.32 ± 2.18

21.13 ± 1.68

27.17 ± 1.10

9k

21.91 ± 1.08

21.73 ± 0.80

27.62 ± 0.91

10k

21.93 ± 1.18

\(\underline{21.64 \pm 0.82}\)

\(\underline{27.80 \pm 0.70}\)

2A

Medical

0k

45.45 ± 2.27

45.66 ± 1.74

53.36 ± 0.97

1k

46.69 ± 3.52

46.60 ± 2.91

54.69 ± 1.74

2k

46.97 ± 1.56

46.22 ± 1.86

54.80 ± 1.69

3k

43.50 ± 2.89

43.47 ± 2.68

52.46 ± 2.42

4k

45.80 ± 2.97

45.21 ± 2.48

53.98 ± 2.00

5k

45.32 ± 2.04

44.83 ± 1.46

53.98 ± 1.10

6k

46.98 ± 2.46

46.05 ± 2.35

54.34 ± 2.12

7k

45.94 ± 3.08

45.36 ± 2.57

54.42 ± 1.27

8k

\(\underline{46.98 \pm 2.65}\)

\(\underline{46.28 \pm 2.26}\)

\(\underline{54.82 \pm 1.83}\)

9k

46.15 ± 1.78

45.85 ± 2.06

55.01 ± 1.06

10k

44.90 ± 2.34

43.92 ± 1.72

53.40 ± 1.46

2B

Music

0k

46.88 ± 4.28

47.27 ± 3.59

53.97 ± 2.48

1k

52.94 ± 2.05

52.92 ± 2.37

58.59 ± 2.05

2k

51.13 ± 2.99

51.91 ± 2.39

58.37 ± 1.79

3k

47.29 ± 4.56

47.95 ± 4.17

55.36 ± 2.78

4k

50.98 ± 3.89

51.67 ± 3.55

57.64 ± 2.48

5k

51.45 ± 2.99

51.69 ± 2.47

58.25 ± 1.51

6k

\(\underline{52.24 \pm 4.52}\)

\(\underline{52.25 \pm 3.94}\)

\(\underline{58.49 \pm 2.41}\)

7k

51.28 ± 4.33

51.27 ± 3.25

57.46 ± 1.96

8k

51.76 ± 3.20

52.17 ± 2.01

58.20 ± 1.29

9k

49.46 ± 6.50

49.79 ± 5.01

56.87 ± 3.42

10k

48.86 ± 5.33

49.52 ± 4.61

56.04 ± 2.52

  1. 0k indicates the BERT model. Bold represents the best performance across the comparison models, and underline represents second place