Detecting the community structure and activity patterns of temporal networks: a non-negative tensor factorization...

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Temporal network data is increasingly available, necessitating methods to analyze its mesoscopic structures and temporal dynamics.
• Existing methods for static networks struggle to capture the intertwined topological and activity patterns in time-varying systems.

Purpose of the Study:

• To introduce and evaluate non-negative tensor factorization (NTF) for extracting community-activity structures in temporal networks.
• To demonstrate NTF's capability in simultaneously identifying communities and tracking their temporal activity.

Main Methods:

• Representing temporal networks as three-way tensors (nodes, nodes, time).
• Applying non-negative tensor factorization to decompose the tensor into community and activity components.
• Utilizing quality metrics for tuning the factorization complexity.

Main Results:

• NTF successfully extracts community-activity structures from temporal network data.
• The method accurately recovers known community structures, validated by school class data.
• Extracted components correlate with known temporal activity patterns and schedules.

Conclusions:

• Non-negative tensor factorization is a powerful, intrinsically temporal method for analyzing complex network dynamics.
• This approach effectively bridges the gap between static network analysis and the study of time-varying systems.
• NTF offers a robust framework for uncovering functional relationships within temporal networks.