Skip to main content

Table 5 Results of the fitting of BCCM to five real-world graphs, with vertex blocks given obtained from five different community detection algorithms

From: The block-constrained configuration model

Data specifications

dataset

vertices

edges

directed

self-loops

 

rfid

75

32424

False

False

 

karate

34

231

False

False

 

UKfaculty

81

3730

True

False

 

USairports

755

23473

True

True

 

enron

184

125409

True

True

 

Number of clusters

dataset

fast_greedy

infomap

label_prop

spinglass

louvain

rfid

6

4

3

7

6

karate

3

3

3

4

4

UKfaculty

5

10

7

7

5

USairports

28

57

40

NA

21

enron

11

22

20

NA

10

\(\bf \Delta _{i}^{\text {AIC}{}}\)

dataset

fast_greedy

infomap

label_prop

spinglass

louvain

rfid

1856

12370

13523

0

1856

karate

28

28

28

4

0

UKfaculty

992

0

960

523

992

USairports

1903

2759

5133

NA

0

enron

0

9881

46945

NA

1956

\(\bf \Delta _{i}^{\text {BIC}{}}\)

dataset

fast_greedy

infomap

label_prop

spinglass

louvain

rfid

1798

12219

13339

0

1798

karate

14

14

14

4

0

UKfaculty

743

0

792

355

743

USairports

3315

14227

9883

NA

0

enron

0

11702

48347

NA

1849

  1. The first table reports information about the five different graphs used. The second table reports the number of clusters detected by each algorithm for each dataset. The algorithm detecting the smallest number of clusters is highlighted in bold, and the algorithm detecting the largest number of clusters is highlighted in italic. The third table reports AIC differences of the different models computed using the different vertex blocks. The fourth table reports BIC differences of the different models computed using the different vertex blocks. The best model, i.e., the one with the lowest AIC/BIC score, respectively, is highlighted in bold. Because the spin-glass algorithm is not suitable for disconnected graphs, no result is reported for this method for the last two real-world graphs