Block-corrected Modularity for Community Detection

2025-02-27

AI Summary

In complex networks, observed node attributes (e.g., publication year, institutional affiliation) may induce spurious groupings that obscure the true community structure driven by latent, unobserved attributes. Method: We propose Block-Corrected Modularity—a theoretically grounded modularity variant that explicitly nullifies the influence of given block structures (e.g., temporal or institutional partitions), thereby enabling unbiased detection of communities driven by unknown attributes. Our approach integrates spectral optimization with two Louvain-inspired refinement heuristics and constructs a temporally corrected citation network using OpenAlex data. Results: On diverse synthetic benchmarks, our method accurately recovers ground-truth communities. Applied to real academic citation networks, it effectively disentangles temporal confounding effects and reveals temporally stable research-topic communities—outperforming conventional null-model-based approaches in both accuracy and interpretability.

Technology Category

• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Reasoning under Uncertainty: Causality

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Social Networks and Social Media: Computational social science
• Web Mining and Content Analysis: Community question answering

Abstract

Unknown node attributes in complex networks may introduce community structures that are important to distinguish from those driven by known attributes. We propose a block-corrected modularity that discounts given block structures present in the network to reveal communities masked by them. We show analytically how the proposed modularity finds the community structure driven by an unknown attribute in a simple network model. Further, we observe that the block-corrected modularity finds the underlying community structure on a number of simple synthetic network models while methods using different null models fail. We develop an efficient spectral method as well as two Louvain-inspired fine-tuning algorithms to maximize the proposed modularity and demonstrate their performance on several synthetic network models. Finally, we assess our methodology on various real-world citation networks built using the OpenAlex data by correcting for the temporal citation patterns.

Problem

• Detect community structures in networks
• Discount known block structures
• Reveal communities masked by attributes

Innovation

• Block-corrected modularity
• Spectral method
• Louvain-inspired algorithms