Finding Statistically Significant Communities in Networks
Figure 8
Test on undirected and unweighted LFR benchmark with overlapping communities.
The parameters are: ,
,
,
,
. S and B indicate the usual ranges of community sizes we use:
and
, respectively. We tested OSLOM against two recent methods to find covers in graphs: COPRA [52] and MOSES [54]. The left panel displays the normalized mutual information (NMI) between the planted cover and the one recovered by the algorithm, as a function of the fraction of overlapping vertices. Each overlapping vertex is shared between two clusters. The four curves correspond to different values of the mixing parameter
(
and
) and to the community size ranges S and B. The right panel shows a test on graphs whose vertices are all shared between clusters. Each vertex is member of the same number of clusters. The plot shows the NMI as a function of the number of memberships of the vertices. Each curve corresponds to a given value of the average degree
. The graph parameters are
,
,
,
,
. Community sizes are in the range
.