Graphlet- and motif-based link prediction in large networks.

Area of Science:

• Network Science
• Computational Biology
• Data Mining

Background:

• Link prediction is crucial for understanding biological, social, and technological networks.
• Current methods often fail to scale, require specific node data, or overlook complex network structures.

Purpose of the Study:

• To introduce BLANT-Predict, a novel topology-only framework for link prediction.
• To evaluate BLANT-Predict's performance against established methods on diverse real-world networks.

Main Methods:

• Utilizes sampled graphlets and orbit-pair frequencies for edge ranking.
• Employs a topology-only approach, eliminating the need for domain-specific node attributes.
• Tests scalability on networks up to 1 million nodes and 3 million edges.

Main Results:

• BLANT-Predict demonstrates superior precision compared to 13 baseline methods across 12 real-world networks.
• The framework exhibits strong scalability, handling large-scale network data efficiently.
• Maintains high precision when predicting future network links, outperforming standard cross-validation.

Conclusions:

• BLANT-Predict offers a practical and effective solution for link prediction, especially in large and dynamic networks.
• Its topology-only nature and scalability make it broadly applicable across various network domains.
• The model's ability to predict future links highlights its real-world deployment value.