Skip to main content

Table 3 Spearman correlation between other measures and AUC

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

NDCG

Amazon

0.1%

34%

11%

-35%

54%

68%

79%

88%

Amazon

1%

16%

-5%

51%

1%

96%

45%

89%

Amazon

10%

-34%

28%

0%

28%

92%

38%

78%

Amazon

25%

-30%

18%

-49%

74%

84%

33%

92%

Amazon

50%

-42%

17%

35%

71%

83%

58%

92%

DBLP

0.1%

7%

-47%

-50%

26%

52%

42%

91%

DBLP

1%

46%

8%

-22%

-85%

28%

85%

84%

DBLP

10%

2%

-23%

15%

-55%

-36%

88%

32%

DBLP

25%

4%

-28%

5%

-35%

-23%

79%

16%

DBLP

50%

-5%

-25%

12%

-5%

-12%

43%

68%

BlockModel

0.1%

12%

42%

21%

-6%

50%

49%

80%

BlockModel

1%

-8%

29%

23%

-35%

8%

6%

95%

BlockModel

10%

-18%

38%

37%

-40%

40%

47%

99%

BlockModel

25%

-63%

70%

35%

-77%

-23%

3%

100%

BlockModel

50%

-70%

72%

73%

-79%

-22%

0%

-25%

CiteSeer

0.1%

56%

40%

45%

63%

39%

63%

82%

CiteSeer

1%

19%

41%

47%

75%

63%

75%

63%

CiteSeer

10%

70%

55%

69%

52%

13%

14%

93%

CiteSeer

25%

42%

34%

48%

25%

9%

18%

82%

CiteSeer

50%

14%

-7%

22%

41%

35%

47%

78%

PubMed

0.1%

48%

-8%

44%

70%

-10%

-33%

93%

PubMed

1%

47%

27%

46%

37%

-52%

-53%

78%

PubMed

10%

69%

-22%

78%

38%

-9%

-14%

81%

PubMed

25%

54%

-22%

77%

18%

21%

20%

84%

PubMed

50%

13%

7%

25%

-5%

18%

17%

80%

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