Link prediction in real-world multiplex networks via layer reconstruction method.

Area of Science:

• Complex Systems Science
• Network Science
• Data Science

Background:

• Networks model complex systems, but real-world systems often feature diverse connection types.
• Multiplex networks capture this complexity, with nodes connected across different layers and link types.
• Link prediction, crucial for understanding network evolution, traditionally focused on single-layer networks.

Purpose of the Study:

• To investigate the source of performance enhancement in link prediction for multiplex networks.
• To identify the key factors contributing to improved prediction accuracy when utilizing information from multiple network layers.
• To understand the relationship between structural similarity and the difficulty of link prediction in multiplex networks.

Main Methods:

• Proposed a layer reconstruction method to analyze multiplex network structures.
• Conducted experiments on real-world multiplex networks from various disciplines.
• Utilized eigenvector similarity to quantify structural relationships between nodes across layers.

Main Results:

• Demonstrated that eigenvector similarity is a major source of enhancement for link prediction in multiplex networks.
• Showcased the effectiveness of the layer reconstruction method in identifying this enhancement.
• Characterized the impact of low structural similarity on prediction performance, highlighting challenging cases.

Conclusions:

• Eigenvector similarity is a critical factor driving the success of link prediction in multiplex networks.
• The layer reconstruction method provides insights into the structural underpinnings of multiplex network link prediction.
• Understanding structural similarity is key to addressing limitations and improving link prediction in complex, multi-layered systems.