An information theoretic approach to link prediction in multiplex networks

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Real-world networks often exhibit multiple layers of connections.
• Link prediction aims to identify missing connections within these networks.
• Existing methods may not fully leverage inter-layer correlations.

Purpose of the Study:

• To develop a general-purpose, similarity-based method for link prediction in multiplex networks.
• To quantify connection uncertainty using observed inter-layer correlations.
• To enhance prediction accuracy by incorporating link overlap across layers.

Main Methods:

• A novel similarity-based method named SimBins was devised.
• SimBins quantifies connection uncertainty based on inter-layer correlations.
• The method incorporates link overlap effects across network layers.

Main Results:

• SimBins demonstrated superior performance compared to baseline and state-of-the-art methods in link prediction tasks.
• The method showed effectiveness across diverse real-world multiplex networks (social, biological, technological).
• A positive correlation between connection probability in one layer and similarity in others was observed.

Conclusions:

• SimBins offers an effective and generalizable approach for multiplex link prediction.
• The method imposes minimal computational overhead, making it suitable for large-scale networks.
• SimBins advances the field by leveraging inter-layer similarities for improved link discovery.