Dynamic-n-Static Multiplex Graph Representation Learning for Improved Link Prediction.

Area of Science:

• Graph Neural Networks
• Network Science
• Machine Learning

Background:

• Social systems feature diverse relations with unique temporal dynamics, often modeled as temporal multiplex graphs.
• Dynamic-n-static multiplex graphs combine evolving dynamic layers with stable static layers, presenting unique challenges for link prediction.

Purpose of the Study:

• To propose DS-MGN, a representation learning model for dynamic-n-static multiplex graphs.
• To enhance link prediction accuracy by utilizing information from static graph layers.

Main Methods:

• Developed DS-MGN, incorporating a cross-layer neighbor encoding (CLNE) scheme.
• CLNE injects topological priors from static layers to enrich dynamic layer representations.
• Created a new dynamic-n-static multiplex graph dataset for research.

Main Results:

• DS-MGN achieved state-of-the-art performance against 12 baselines on benchmark and custom datasets.
• Demonstrated substantial performance improvements through the CLNE scheme.
• Highlighted the model's scalability and effectiveness.

Conclusions:

• DS-MGN offers a robust solution for link prediction in complex dynamic-n-static multiplex graphs.
• The CLNE scheme is adaptable and enhances existing neural network architectures.
• The model shows promise for real-world applications like social network analysis and recommendation systems.