New Methods for Generating Synthetic Equivalents of Real Social Networks

Abstract

In both empirical and theoretical contexts, we may need to generate synthetic versions of large-scale social networks. Extant network generation models and algorithms do not replicate in full measure the structural complexities of real social networks. While extant approaches can replicate specific network characteristics well (e.g., degree distribution), they fail to replicate other aspects such as presence of overlapping cliques and community structures in social networks. Inspired by some recent developments in community detection algorithms in social networks, we propose two new network generation algorithms that explicitly incorporate the distribution of overlapping communities. We show that our proposed algorithms are measurably superior to the extant algorithms in generating synthetic social networks that more closely resemble actual social networks along a multiplicity of metrics that are used to characterize such networks.