Abstract

To find interesting structure in networks, community detection algorithms have to consider not only the network topology, but also the dynamics of interactions between nodes. We investigate this claim using the paradigm of synchronization in a network of coupled oscillators. As the network evolves to a global equilibrium, nodes belonging to the same community synchronize faster than nodes belonging to different communities. We classify interactions as conservative (eg, random walk) and non-conservative (eg, viral contagion, information diffusion) and formulate a new model of non-conservative interactions. To find multi-scale community structure, we define a similarity function that measures the degree to which nodes are synchronized and use it to hierarchically cluster nodes. We study three data sets, that include a benchmark network, a synthetic graph with a known hierarchical community structure, and a large network of a social media provider. We find that conservative and nonconservative interaction models lead to dramatically different communities, with the non-conservative model revealing communities closer to the ground truth. Our method uncovers a significantly more complex multi-scale organization of networks than previously thought. The discovered structure of a real-world network resembles an onion: in each layer of the hierarchy, we find a large core and a number of small components with a long-tailed size distribution. Our work offers a novel, process-dependent perspective on community detection in real-world social networks.

Date

January 15, 2026

Authors

Rumi Ghosh, Kristina Lerman

Journal

CoRR