Dynamic Infinite Mixed-Membership Stochastic Blockmodel.

Area of Science:

• Social network analysis
• Statistical modeling
• Machine learning

Background:

• Mixed-membership stochastic blockmodels (MMSB) are foundational for social network analysis.
• Existing models struggle with dynamic networks and an unknown number of communities.
• Modeling temporal dynamics and community persistence is crucial for understanding evolving networks.

Purpose of the Study:

• To propose a dynamic infinite mixed-membership stochastic blockmodel (MMSB) for analyzing networks over time.
• To extend existing MMSB frameworks to accommodate potentially infinite communities.
• To introduce parameters for capturing temporal persistence in community memberships.

Main Methods:

• Developed a generalized dynamic infinite MMSB framework.
• Proposed mixture time variant and mixture time invariant models to capture different temporal correlations.
• Implemented two posterior sampling strategies for model inference.
• Validated the models using synthetic and real-world network data.

Main Results:

• The proposed dynamic infinite MMSB framework effectively models evolving social networks.
• The mixture time variant and invariant models successfully depict distinct temporal correlation structures.
• Posterior sampling strategies provided effective inference for the dynamic network models.
• The framework demonstrates applicability on both simulated and empirical network data.

Conclusions:

• The dynamic infinite MMSB offers a robust approach for modeling complex network dynamics.
• The model's ability to handle an unknown and potentially infinite number of communities enhances its flexibility.
• This work provides a valuable tool for understanding temporal changes and membership persistence in social networks.