Figure 1
From: Exploring the limits of community detection strategies in complex networks

Best algorithms in LFR closed benchmarks.
The six algorithms able to recover the initial partition when C ≥ 5% are shown. In these diagrams, the x-axis shows the conversion percentage and the y-axis, the VI value. The red line indicates the VI values obtained when the algorithm solution is compared with the initial structure and the black line, the same comparison, but with the final structure. A perfect identity corresponds to the value VI = 0. Comparing the (VIIE + VIEF)/2 values (blue line) and the VIIF/2 values (dotted line, often invisible, being just below the blue one), we can conclude that Infomap, RB and LPA achieve optimal values until C is very close to 50%. MCL, RN and CPM work accurately only in the easiest analyses (both ends of the benchmark).