Sampling Networks from Modular Compression of Network Flows

Christopher Blöcker, Jelena Smiljanić, Martin Rosvall, Ingo Scholtes

Keywords: flow community, network model, benchmark

TL;DR: We turn the information-theoretic node similarity measure known as "map equation similarity" into a generative network model.

Abstract: Traditional benchmark models generate networks with given structural characteristics such as degree distribution, degree correlations, or community structure but do not consider dynamic processes on networks. Since dynamics are often superordinate to structure and guide network formation, we can learn about the structure from dynamics but lack methods for translating modular dynamics into network structure. To bridge this gap, we introduce a generative network model rooted in the modular compression of dynamic processes on a network as provided by the map equation, an information-theoretic method for community detection. We evaluate our approach by sampling networks according to the modular compression of network flows in empirical networks from different domains. We recover the original community structure and preserve the nodes' expected out-degrees, enabling benchmark networks by sampling from dynamic processes.

Submission Type: Extended abstract (max 4 main pages).

Agreement: Check this if you are okay with being contacted to participate in an anonymous survey.

Submission Number: 52