A robust Bayesian latent position approach for community detection in networks with continuous attributes

Area of Science:

• Network analysis
• Statistical modeling
• Machine learning

Background:

• Multiplex networks are increasingly common, requiring analysis of inter-layer dependencies.
• Community detection is a fundamental task in network analysis.
• Existing models often overlook nodal attributes and complex layer interactions.

Purpose of the Study:

• To propose a full Bayesian mixture model for community detection in single and multi-layer networks.
• To jointly model nodal attributes as a spatial process within a latent space.
• To account for varying dependency strengths across layers in multiplex networks.

Main Methods:

• A full Bayesian mixture model incorporating nodal attributes.
• Joint modeling of latent positions and attribute data.
• Layer-specific dependency factors within a Gaussian mixture prior structure.

Main Results:

• The proposed model outperforms existing benchmark models in simulated examples.
• Demonstrated significant robustness in handling datasets with missing values.
• Successfully applied to a real-world three-layer employee network.

Conclusions:

• The Bayesian mixture model provides a robust framework for community detection in multiplex networks.
• Jointly modeling nodal attributes enhances network analysis accuracy.
• The model's performance indicates its utility for complex, real-world network data.