Generalized L-Modularity for Community Detection Beyond Simple Temporal Networks

2026-05-23

AI Summary

Existing dynamic community detection methods often oversimplify complex temporal networks—characterized by transient events, sustained contacts, delayed interactions, directionality, edge weights, and heterogeneous node types—thereby discarding critical structural information. This work proposes a unified framework that generalizes longitudinal modularity (L-modularity) for the first time to accommodate multimodal interactions, directed edges, weighted ties, and heterogeneous node types. Building upon this generalized formulation, we design a dynamic community detection algorithm that operates directly on the original temporal network without requiring time-window aggregation or multipartite graph projection. Experiments on three real-world datasets demonstrate that our approach effectively uncovers meaningful community structures and significantly enhances both modeling capacity and robustness in capturing complex temporal interactions.

Abstract

Detecting communities in networks is essential for understanding the mesoscopic organization of complex systems. Interactions in most real-world networks evolve over time and exhibit diverse modalities: instantaneous events, continuous contacts that persist over intervals, and delayed interactions where source and destination are temporally separated, as observed in transportation processes. Additionally, interactions may be directed, weighted, or involve multiple node types. Existing methods for community detection in temporal networks typically handle only limited subsets of these features. When applied to real-world data, they often rely on simplifying transformations, such as aggregating interactions into time windows, projecting multipartite structures onto unipartite graphs, or ignoring edge directions and weights, leading to a loss of information. In this work, we generalize Longitudinal Modularity (L-Modularity) and the LAGO algorithm into a unified framework for dynamic community detection in complex link streams. Experiments on three real-world datasets demonstrate that our approach discovers meaningful communities in temporal networks with diverse interaction types.

Problem

• community detection
• temporal networks
• complex link streams
• interaction heterogeneity
• information loss

Innovation

• Generalized L-Modularity
• dynamic community detection
• complex link streams
• temporal networks
• LAGO algorithm