CoDeSEG algorithm achieves state-of-the-art large-scale community detection via entropy game

Community detection in large-scale networks—graph structures with millions of nodes and billions of edges—remains computationally challenging. Existing methods often falter on efficiency, restrict themselves to specific graph types (e.g., unweighted or undirected), or only handle static communities. To solve this, researchers from multiple institutions propose CoDeSEG (Community Detection via Structural Entropy Game). The algorithm reimagines community detection as a potential game where nodes decide to stay or move based on a utility function that minimizes the network’s two-dimensional structural entropy. This game-theoretic approach allows CoDeSEG to efficiently identify densely interconnected clusters without sacrificing accuracy.

Beyond static detection, CoDeSEG introduces two key innovations: a structural entropy–based node overlapping heuristic that detects overlapping communities with near-linear complexity, and a cascading influence propagation–based adaptive update strategy that extends detection to dynamic graphs—identifying nodes whose affiliations change as the network evolves. Tested on 14 diverse large-scale networks, CoDeSEG achieves state-of-the-art performance across all three tasks (overlapping, non-overlapping, dynamic), with substantial efficiency gains over prior work. The paper, published on arXiv (2607.13713), opens the door to practical, real-time community detection in social networks, bioinformatics, and beyond.