Inference of dynamic hypergraph representations in temporal interaction data

Area of Science:

• Complex Systems Science
• Data Science
• Network Science

Background:

• Many scientific domains generate time-stamped interaction data between distinct item categories (e.g., users and products, insects and plants).
• These datasets can be represented as temporal hypergraphs, but choosing optimal time windows for snapshots is challenging and impacts analysis.
• Existing methods lack a principled approach for determining the number and duration of temporal snapshots.

Purpose of the Study:

• To develop a principled, data-driven method for extracting optimal temporal hypergraph snapshots from event data.
• To address the challenge of selecting appropriate time windows for hypergraph representation of temporal interactions.
• To enhance the analysis of structural regularities in time-varying interaction datasets.

Main Methods:

• Proposed a nonparametric solution based on the minimum description length (MDL) principle.
• Developed a method to extract temporal hypergraph snapshots that optimally capture structural regularities.
• Applied and validated the method on both synthetic and real-world datasets, including human mobility data.

Main Results:

• The proposed method successfully recovers planted hypergraph structures from noisy data.
• Demonstrated the ability to reveal meaningful fluctuations in human mobility patterns.
• The MDL-based approach provides an optimal and data-driven strategy for temporal hypergraph construction.

Conclusions:

• The developed method offers a robust solution for modeling temporal event data as hypergraphs.
• Optimizing temporal snapshot extraction enhances the discovery of network structures and dynamics.
• This approach has broad applicability across social and natural sciences for analyzing complex interactions.