The block-constrained configuration model

Applied Network Science

Table 4 Comparison of the goodness-of-fit of 5 different block structures detected by five different community detection algorithms

	fast_greedy	infomap	label_prop	spinglass	louvain	original
B	2	4	2	7	2	3
\(\Delta _{i}^{\text {AIC}{}}\)	4	0	4	40	4	-282
w_i	0.12	0.88	0.12	0	0.12
B	2	4	2	7	2	3
\(\Delta _{i}^{\text {BIC}{}}\)	0	57	0	251	0	-260
w_i	1	0	1	0	1

The different partitions are compared in terms of the AIC and BIC obtained by the corresponding BCCM. Note that model weights are computed normalizing over three models because the three algorithms that detect two blocks detect the same partitioning. In the right-most column, we give the results corresponding to the ground-truth block partitioning. Because none of the algorithms was able to detect the original partitioning, we do not include the corresponding model when computing weights