Table 2 Comparison of the NMIs accuracy of different methods on 6 Stanford large networks with ground-truth of overlapping communities22. Here, n is the number of nodes, m the number of links and c the number of communities. M denotes a million and k one thousand. The mean values and standard deviations are based on 50 runs and Louvain, LC and CPM are deterministic algorithms without standard deviations. The larger the NMI of an overall community structure, the better the structure is. The best NMIs for these networks are shown in bold and underlined
From: Identification of hybrid node and link communities in complex networks
|  |  |  |  |         Methods | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
Datasets/NMIs (%) | n | m | c | FUA (node) | LC (link) | CPM (overlap) | NModel (node) | LModel (link) | NLC-EM (hybrid) | NLC-NMF (hybrid) |
LiveJournal | 4.0M | 34.9M | 310k | 20.07 | 14.77 | 18.84 | 27.64 ± 0.56 | 23.69 ± 0.48 | 28.74 ± 0.49 | 41.02 ± 1.15 |
Friendster | 120M | 2,600M | 1.5M | 28.65 | 17.18 | 27.59 | 32.82 ± 1.07 | 32.36 ± 0.57 | 38.97 ± 0.51 | 23.50 ± 0.62 |
Orkut | 3.1M | 120M | 8.5M | 25.60 | 17.73 | 26.54 | 26.90 ± 0.55 | 23.69 ± 0.43 | 28.59 ± 0.40 | 33.83 ± 0.24 |
Youtube | 1.1M | 3.0M | 30k | 24.06 | 17.81 | 13.80 | 17.82 ± 0.60 | 29.91 ± 0.60 | 33.92 ± 0.46 | 31.70 ± 0.14 |
DBLP | 0.43M | 1.3M | 2.5k | 16.83 | 14.12 | 17.99 | 15.20 ± 0.51 | 13.71 ± 0.52 | 14.98 ± 0.30 | 35.49 ± 0.36 |
Amazon | 0.34M | 0.93M | 49k | 24.73 | 17.56 | 18.10 | 26.70 ± 0.40 | 24.74 ± 0.70 | 29.24 ± 0.59 | 41.44 ± 0.62 |
