Ego-zones: non-symmetric dependencies reveal network groups with large and dense overlaps

Applied Network Science

Table 17 Zones in networks with ground-truth communities

Network	Communities			Matching zones			MCC	MCC = 1		MCC >0.9
	Count	Min	Max	Count	Percent	Avg size	Avg	Percent	Avg size	Percent	Avg size
anobii	4797	1	7062	4560	95.06	2.502	0.337	3.49	1.119	3.51	1.150
com-amazon	5000	3	328	1442	28.84	9.549	0.853	44.52	6.467	57.49	7.943
com-dblp	5000	6	7556	4850	97.00	8.721	0.869	39.75	7.487	55.53	7.810
com-youtube	5000	2	2217	4368	87.36	5.485	0.648	18.10	2.217	18.407	2.281
LFR _{20 500 2000}	186	7	500	175	94.09	25.514	0.703	5.71	8.000	16.57	9.690
LFR _{7 60 4000}	1000	3	60	928	92.80	8.179	0.628	3.66	3.147	4.85	3.822

There is a low match of zones and corresponding communities for the anobii network. On the other hand, for the com-amazon and com-dblp networks, the match is high. For anobii and both the LFR networks, only a low proportion of well-matching zones exists; the average size of the matching zones is low for anobii and high for the first LFR Network. Table 16 shows, however, that the match for the com-amazon and LFR networks is much better than for non-overlapping communities detected by the Louvain algorithm. Conversely, for the anobii and com-youtube networks, the match for non-overlapping communities is higher