Modularity in Complex Multilayer Networks with Multiple Aspects: A Static Perspective

Han Zhang, Chang-Dong Wang, Jian-Huang Lai, Philip S. YuarXiv 2016 · arXiv

Introduction

In modern network science, real-world systems are increasingly modeled as complex multilayer structures rather than simple graphs. Whether analyzing social interactions across multiple digital platforms or mapping brain activity across diverse subjects, these systems often possess multiple "aspects"—dimensions such as time, relationship type, or data source. Detecting communities within these intricate structures remains a significant challenge. This paper introduces a novel static framework for defining modularity in multi-aspect multilayer networks, moving away from traditional dynamic random-walk models toward a more robust structural perspective.

The Challenge of Multiple Aspects

Traditional community detection methods often rely on a dynamic perspective, where communities are defined by the trapping of a "random walker." However, this approach can become mathematically cumbersome or conceptually misaligned when dealing with temporal networks or layers without a natural flow. Furthermore, most existing tools struggle with networks characterized by multiple aspects—for instance, a network where one aspect represents interaction types (e.g., email vs. phone) and another represents time slices.

A conceptual representation of a multilayer network with two aspects (I and II), showing intra-layer edges and inter-layer couplings.

The research addresses this gap by proposing a metric that is purely structural (static) and flexible enough to handle multi-dimensional aspects simultaneously.

Methodology: A Static Hamiltonian Perspective

The authors provide two primary contributions to the field:

  1. Generalized Static Modularity: Derived from the Hamiltonian of a Potts model in statistical mechanics, this "static" view defines a community as a group of nodes maximizing internal cohesion. This is measured through dense internal edges and strong "couplings" between a node and its counterparts in other layers.
  2. mSpec (Multilayer Spectral Optimization): To optimize community detection, the authors developed a spectral bisection algorithm. The entire multilayer system is represented as a "supra-adjacency matrix"—a flattened representation of the entire network. By utilizing the dominant eigenvector of this matrix, the algorithm iteratively splits the network into communities. This provides a deterministic and stable alternative to heuristic methods like the Louvain algorithm.

Experimental Validation

The framework was validated across a diverse range of synthetic and empirical datasets:

  • Synthetic Benchmarks: A multilayer adaptation of the Zachary Karate Club network.
  • Empirical Networks: Seven datasets including CKM-Physicians Innovation (drug adoption), London Transportation (train station connectivity), and CS-Aarhus (social relationships).
  • EEG Case Study: A complex two-aspect network involving 128 electrodes across 11 test subjects.

Key Results and Insights

The proposed mSpec algorithm demonstrated significant improvements over existing methods:

  • Superior Performance: Across seven empirical datasets, mSpec achieved an average of 18.65% higher modularity than competing methods.
  • Stability and Consistency: Unlike heuristic methods that can produce varying results across runs, the spectral approach proved far less sensitive to noise or random initialization.
  • Biological Relevance: In the EEG case study, the algorithm successfully identified communities corresponding to known functional brain regions, such as the visual and prefrontal cortices, during visual stimuli tests.

Conclusion

By shifting the perspective from dynamic random walks to a static energy-based model, this research provides a more intuitive and stable metric for community detection in complex networks. While the spectral method is computationally intensive for massive datasets, its reliability and ability to handle "unevenly distributed views" make it a powerful tool for domain-specific applications like neuroscience. Future research will focus on automating coupling parameter selection and enhancing scalability for even larger datasets.

Summary updated

Abstract

Complex systems are usually illustrated by networks which captures the topology of the interactions between the entities. To better understand the roles played by the entities in the system one needs to uncover the underlying community structure of the system. In recent years, systems with interactions that have various types or can change over time between the entities have attracted an increasing research attention. However, algorithms aiming to solve the key problem - community detection - in multilayer networks are still limited. In this work, we first introduce the multilayer network model representation with multiple aspects, which is flexible to a variety of networks. Then based on this model, we naturally derive the multilayer modularity - a widely adopted objective function of community detection in networks - from a static perspective as an evaluation metric to evaluate the quality of the communities detected in multilayer networks. It enables us to better understand the essence of the modularity by pointing out the specific kind of communities that will lead to a high modularity score. We also propose a spectral method called mSpec for the optimization of the proposed modularity function based on the supra-adjacency representation of the multilayer networks. Experiments on the electroencephalograph network and the comparison results on several empirical multilayer networks demonstrate the feasibility and reliable performance of the proposed method.
BibTeX
@misc{zhang2016modularity,
  author = {Han Zhang and Chang-Dong Wang and Jian-Huang Lai and Philip S. Yu},
  title = {{Modularity in Complex Multilayer Networks with Multiple Aspects: A Static Perspective}},
  year = {2016},
  eprint = {1605.06190},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1605.06190},
}
Figure