[Paper Review] Temporal Network Comparison using Graphlet-orbit Transitions

David Aparício, Pedro Ribeiro+1 more|arXiv (Cornell University)|Jul 14, 2017

Complex Network Analysis Techniques1 references3 citations

TL;DR

This paper proposes a temporal network comparison method using graphlet-orbit transitions, which captures dynamic structural changes by tracking how 4-node subgraph patterns evolve over time. The orbit-transition-agreement (OTA) metric outperforms static methods in grouping networks by category and reveals interpretable, system-specific evolution patterns such as stable collaboration groups versus transient physical interactions.

ABSTRACT

Networks are widely used to model real-world systems and uncover their topological features. Network properties such as the degree distribution and shortest path length have been computed in numerous real-world networks, and most of them have been shown to be both scale-free and small-world networks. Graphlets and network motifs are subgraph patterns that capture richer structural information than aforementioned global network properties, and these local features are often used for network comparison. However, past work on graphlets and network motifs is almost exclusively applicable only for static networks. Many systems are better represented as temporal networks which depict not only how a system was at a given stage but also how they evolved. Time-dependent information is crucial in temporal networks and, by disregarding that data, static methods can not achieve the best possible results. This paper introduces an extension of graphlets for temporal networks. Our proposed method enumerates all 4-node graphlet-orbits in each network-snapshot, building the corresponding orbit-transition matrix in the process. Our hypothesis is that networks representing similar systems have characteristic orbit transitions which better identify them than simple static patterns, and this is assessed on a set of real temporal networks split into categories. In order to perform temporal network comparison we put forward an orbit-transition-agreement metric (OTA). OTA correctly groups a set of temporal networks that both static network motifs and graphlets fail to do so adequately. Furthermore, our method produces interpretable results which we use to uncover characteristic orbit transitions, and that can be regarded as a network-fingerprint.

Motivation & Objective

• To address the limitation of static network analysis in capturing temporal dynamics of real-world systems.
• To develop a method that leverages local, subgraph-level structural patterns to compare evolving networks more effectively than global metrics or static graphlets.
• To create an interpretable, fingerprint-like representation of network evolution through orbit-transition matrices.
• To demonstrate that temporal evolution patterns—specifically transitions between graphlet-orbits—can better distinguish network categories than static motifs or graphlets.

Proposed method

• Enumerates all 4-node graphlet-orbits in each network snapshot to capture detailed local topology.
• Constructs orbit-transition matrices that track changes in orbit configurations across consecutive time steps.
• Defines the orbit-transition-agreement (OTA) metric to compare similarity between networks based on their transition matrices.
• Discretizes transition frequencies into rare, common, and frequent intervals for visual interpretation and pattern recognition.
• Applies the method to real-world temporal networks across collaboration, physical interaction, crime, and bipartite categories.
• Uses visualizations of orbit-transition fingerprints to interpret and compare network dynamics across categories.

Experimental results

Research questions

• RQ1 Can graphlet-orbit transitions effectively distinguish between different types of real-world temporal networks?
• RQ2 Does incorporating temporal dynamics into graphlet analysis improve network comparison beyond static methods?
• RQ3 Are there characteristic orbit-transition patterns that reflect the underlying functional or structural dynamics of specific network types?
• RQ4 Can the resulting orbit-transition fingerprints provide interpretable insights into network evolution?

Key findings

• The OTA metric successfully grouped temporal networks by category, while static graphlet-degree-agreement and motif-fingerprint distance failed to do so.
• Collaboration networks like Authenticus showed relatively stable orbits, with the square-orbit (O5) being the most unstable, reflecting loose group cohesion.
• Physical interaction networks like Conference exhibited highly unstable orbits, with frequent transitions, indicating transient and fluid interactions.
• In arXiv hep-ph, star-to-clique transitions were common, suggesting physicists form tightly connected groups earlier than in general collaboration networks.
• Orbit-transition fingerprints revealed that networks of the same category shared similar transition profiles, even when individual transition frequencies varied.
• The method produced highly interpretable results, enabling researchers to infer behavioral dynamics such as group formation, dissolution, and stability from transition patterns.