Table 1 Average S_orig and S_max values in the LFR and RC benchmarks. Statistical significance (p) was estimated using a two-tailed Student t test. ns: non-significant differences. In italics, the benchmarks containing quasi-random networks, discarded for the main analyses (summarized in Figures 2 and 3), but included in the analyses shown in Figure 5, below

The performance of the algorithms in the limit cases (μ = 0.7, R = 50%) and beyond those limits (μ = 0.8 – 0.9, R = 60 – 90%) are correlated.

A statistically significant correlation was found, despite the fact that some algorithms, such as Infomap or LPA, totally collapsed. These algorithms established partitions consisting in a single community, which led to VI = 0 when compared with the original distribution.

Full size image

From: Surprise maximization reveals the community structure of complex networks