Skip to main content

Table 4 Spearman correlation between other measures and NDCG

From: Unsupervised evaluation of multiple node ranks by reconstructing local structures

  

Structural Measures

Local Structure Measures

 

Network

Examples

-Cond/nce

Density

Modularity

LinkCC

LinkAUC

HopAUC

AUC

Amazon

0.1%

8%

7%

-1%

78%

34%

91%

88%

Amazon

1%

29%

9%

30%

28%

80%

67%

89%

Amazon

10%

-42%

22%

-2%

59%

67%

-3%

78%

Amazon

25%

-23%

12%

-43%

77%

64%

9%

92%

Amazon

50%

-41%

12%

36%

55%

75%

54%

92%

DBLP

0.1%

-12%

-48%

-27%

25%

64%

55%

91%

DBLP

1%

14%

11%

-12%

9%

39%

66%

84%

DBLP

10%

-5%

20%

-57%

-10%

15%

33%

32%

DBLP

25%

2%

36%

31%

-11%

-11%

-6%

16%

DBLP

50%

-28%

-19%

17%

-29%

39%

49%

68%

BlockModel

0.1%

20%

11%

31%

-9%

20%

23%

80%

BlockModel

1%

-8%

25%

24%

-27%

26%

22%

95%

BlockModel

10%

-16%

35%

38%

-40%

39%

46%

99%

BlockModel

25%

-64%

72%

-78%

-77%

-22%

4%

100%

BlockModel

50%

72%

-52%

20%

62%

75%

75%

-25%

CiteSeer

0.1%

51%

16%

40%

62%

73%

89%

82%

CiteSeer

1%

10%

31%

35%

62%

67%

56%

63%

CiteSeer

10%

64%

52%

64%

54%

27%

26%

93%

CiteSeer

25%

14%

54%

31%

50%

-23%

-3%

82%

CiteSeer

50%

-20%

27%

-9%

79%

10%

44%

78%

PubMed

0.1%

57%

12%

26%

60%

5%

-7%

93%

PubMed

1%

53%

16%

53%

36%

-11%

-8%

78%

PubMed

10%

60%

-39%

78%

33%

31%

26%

81%

PubMed

25%

24%

3%

52%

-6%

32%

30%

84%

PubMed

50%

-15%

32%

-17%

-48%

-11%

-11%

80%

  1. The value unsupervised measure closest to 100% in each setting (bolded) is the one best mimicking NDCG for comparing pairs of ranking algorithms