Link Prediction on Complex Networks: An Experimental Survey.

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Complex networks model real-world relationships, significantly impacted by global events such as COVID-19.
• Link prediction is crucial for analyzing network structures, identifying missing or future connections.
• Existing link prediction methods vary based on network topology and entity properties.

Purpose of the Study:

• To propose a novel taxonomy for link prediction methods.
• To provide a comprehensive overview of current link prediction techniques.
• To investigate network embedding-based methods, including graph neural networks.

Main Methods:

• Developed a new five-category taxonomy for link prediction methods.
• Conducted a comprehensive review of existing link prediction approaches.
• Analyzed 36 diverse datasets representing seven network types.
• Performed extensive experiments using network embedding and graph neural network methods.

Main Results:

• A new classification system for link prediction methods was established.
• Network embedding and graph neural network methods showed promising results.
• Experimental analysis identified suitable link prediction approaches for different network topologies.

Conclusions:

• The proposed taxonomy offers a structured understanding of link prediction.
• The study highlights the effectiveness of network embedding for link prediction.
• Identifying optimal methods for specific network types is essential for accurate analysis.