Role Analysis in Networks using Mixtures of Exponential Random Graph Models

Area of Science:

• Network analysis
• Statistical modeling
• Computational social science

Background:

• Understanding actor roles is crucial in network analysis.
• Existing methods may not fully capture roles defined by local network structures.
• Ego-networks offer a localized perspective on actor behavior and influence.

Purpose of the Study:

• To introduce a novel and flexible framework for identifying actor roles within networks.
• To define roles based on local network connectivity patterns using ego-networks.
• To develop a statistical model for clustering nodes into distinct roles.

Main Methods:

• A mixture of Exponential-family Random Graph Models (ego-ERGM) was developed for ego-networks.
• An Expectation-Maximization algorithm was employed for parameter estimation.
• Maximum pseudo-likelihood approximation was used to infer cluster assignments.

Main Results:

• The ego-ERGM framework successfully clusters nodes into roles based on local connectivity.
• The developed Expectation-Maximization algorithm efficiently estimates model parameters.
• The method demonstrated flexibility and utility on both simulated and real-world network data.

Conclusions:

• The ego-ERGM provides a robust method for role discovery in network analysis.
• This approach enhances the understanding of actor positions and functions within networks.
• The framework is applicable to diverse network types and research questions.