2018/04/05 by A. Roxana Pamfil, Pamfil, A. Roxana, Sam Howison +5 · 2 citations
Physics and Astronomy · Psychology · #Complex Network Analysis Techniques #Data Analysis #FOS: Computer and information sciences #FOS: Mathematics #FOS: Physical sciences #Mental Health Research Topics #Opinion Dynamics and Social Influence #Physics and Society (physics.soc-ph) #Probability (math.PR) #Social and Information Networks (cs.SI) #Statistics and Probability (physics.data-an)
paper · pdf · doi:10.48550/arxiv.1804.01964
openalex publication_date 2018/04/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Characterizing large-scale organization in networks, including multilayer\nnetworks, is one of the most prominent topics in network science and is\nimportant for many applications. One type of mesoscale feature is community\nstructure, in which sets of nodes are densely connected internally but sparsely\nconnected to other dense sets of nodes. Two of the most popular approaches for\ncommunity detection are to maximize an objective function called "modularity"\nand to perform statistical inference using stochastic block models.\nGeneralizing work by Newman on monolayer networks (Physical Review E 94,\n052315), we show in multilayer networks that maximizing modularity is\nequivalent, under certain conditions, to maximizing the posterior probability\nof community assignments under a suitably chosen stochastic block model. We\nderive versions of this equivalence for various types of multilayer structure,\nincluding temporal, multiplex, and multilevel networks. We consider cases in\nwhich the key parameters are constant, as well as ones in which they vary\nacross layers; in the latter case, this yields a novel, layer-weighted version\nof the modularity function. Our results also help address a longstanding\ndifficulty of multilayer modularity-maximization algorithms, which require the\nspecification of two sets of tuning parameters that have been difficult to\nchoose in practice. We show how to perform this parameter selection in a\nstatistically-grounded way, and we demonstrate the effectiveness of our\napproach on both synthetic and empirical networks.\n