[Paper Review] Towards effective visual analytics on multiplex and multilayer networks

Matteo Magnani, Luca Rossi|arXiv (Cornell University)|Jan 7, 2015

Complex Network Analysis Techniques11 references3 citations

TL;DR

This paper proposes a novel visual analytics approach for multiplex networks by leveraging network metrics—specifically relevance and exclusive relevance—to simplify and highlight meaningful structural patterns, reducing visual clutter. By filtering nodes based on these metrics and comparing results to random null models, the method reveals non-random, localized clusters that are not apparent in standard visualizations, demonstrating that metric-driven simplification enhances insight extraction in complex multilayer systems.

ABSTRACT

In this article we discuss visualisation strategies for multiplex networks. Since Moreno's early works on network analysis, visualisation has been one of the main ways to understand networks thanks to its ability to summarise a complex structure into a single representation highlighting multiple properties of the data. However, despite the large renewed interest in the analysis of multiplex networks, no study has proposed specialised visualisation approaches for this context and traditional methods are typically applied instead. In this paper we initiate a critical and structured discussion of this topic, and claim that the development of specific visualisation methods for multiplex networks will be one of the main drivers pushing current research results into daily practice.

Motivation & Objective

• To address the lack of specialized visualization techniques for multiplex networks, which currently rely on generic, single-layer methods.
• To reduce visual clutter in multiplex networks by simplifying the representation using network metrics.
• To identify and highlight non-random, localized structural patterns in multiplex networks that are obscured in standard visualizations.
• To validate that the detected patterns are not random by comparing with randomized null models.
• To demonstrate that simplification based on relevance and exclusive relevance reveals hidden community structures in multilayer networks.

Proposed method

• The method applies relevance and exclusive relevance measures to identify nodes that are highly connected within specific layers of a multiplex network.
• Nodes are filtered based on thresholds of relevance and exclusive relevance, creating simplified subnetworks that preserve only the most structurally significant connections.
• The approach uses a null model where nodes are randomly selected with the same degree as in the real data, preserving edge counts but not structural relevance.
• Transitivity (clustering coefficient) is computed for both real and null-model networks to assess whether detected clusters are statistically significant.
• Visualizations are generated from the filtered networks to emphasize structural patterns, with node sizes and colors indicating metric values.
• The method is applied to a real-world multiplex network with five layers (e.g., work, lunch, etc.), demonstrating its effectiveness on empirical data.

Experimental results

Research questions

• RQ1 Can network metrics such as relevance and exclusive relevance be used to simplify multiplex network visualizations and reduce visual clutter?
• RQ2 Do the simplified networks generated via metric-based filtering reveal non-random, localized structural patterns not visible in full multiplex representations?
• RQ3 Are the detected clusters in the simplified networks statistically significant, or could they arise by chance?
• RQ4 How does the performance of the method vary with different threshold values for relevance and exclusive relevance?
• RQ5 Can the proposed simplification approach be generalized to sets of layers, enabling combinatorial exploration of multilayer structures?

Key findings

• The transitivity (clustering coefficient) of networks simplified using relevance and exclusive relevance is consistently higher than in corresponding random null models, indicating the presence of non-random, localized clusters.
• At high relevance thresholds, transitivity drops in both real and random networks due to network sparsity, but the real networks maintain higher transitivity values, confirming structural significance.
• The method successfully identifies U4 as a central hub across layers, particularly in the work and lunch layers, with strong inter-layer bridging roles.
• The use of exclusive relevance reveals that nodes highly relevant to only one layer tend to form localized, well-connected groups, suggesting layer-specific community structures.
• The results show that visualizing all layers simultaneously leads to information overload, while metric-based simplification enhances interpretability and insight discovery.
• The approach demonstrates that simplification based on network metrics can reveal hidden community structures that are not detectable through standard visualization techniques.