An Introduction to Community Detection in Multi-layered Social Network

Social communities extraction and their dynamics are one of the most important problems in today’s social network analysis. During last few years, many researchers have proposed their own methods for group discovery in social networks. However, almost none of them have noticed that modern social networks are much more complex than few years ago. Due to vast amount of different data about various user activities available in IT systems, it is possible to distinguish the new class of social networks called multi-layered social network. For that reason, the new approach to community detection in the multi-layered social network, which utilizes multi-layered edge clustering coefficient is proposed in the paper.

💡 Research Summary

The paper addresses the increasingly complex nature of modern social networks, which are no longer adequately represented by a single graph but rather by a collection of interrelated layers capturing different types of user interactions (e.g., friendships, follows, comments, likes, messages). While many community‑detection algorithms have been proposed for single‑layer networks, only a few attempts have been made to exploit the full richness of multi‑layered structures, and those attempts typically either collapse all layers into one aggregated graph or treat each layer independently and later merge the results. Both strategies discard valuable cross‑layer information and can lead to distorted community boundaries.

To overcome these limitations, the authors introduce the Multi‑layered Edge Clustering Coefficient (MECC). For any edge (u, v) that appears in one or more layers, MECC quantifies how consistently the two endpoints share common neighbors across all layers. Formally, let Γ⁽ˡ⁾(u) denote the neighbor set of node u in layer l (l = 1…L). The MECC of edge e = (u, v) is defined as

\