Portal Nodes Screening for Large Scale Social Networks

Area of Science:

• Social Network Analysis
• Statistical Modeling
• Data Science

Background:

• Network autoregression models (NAM) are crucial for understanding user behavior on large social networks.
• Identifying influential users (portal nodes) is essential for network analysis and understanding information diffusion.
• Existing methods may not fully capture heterogeneous and sparse network effects.

Purpose of the Study:

• To develop a robust method for identifying influential users within the NAM framework.
• To specifically address models with heterogeneous and sparse network effect coefficients.
• To establish theoretical guarantees for the proposed identification procedure.

Main Methods:

• Utilizing a network autoregression model (NAM) with heterogeneous and sparse coefficients.
• Designing a screening procedure to identify portal nodes based on non-zero network effect coefficients.
• Applying quasi-maximum likelihood estimation to quantify influential powers.
• Establishing asymptotic normality for the proposed estimators.
• Employing a local linear approximation algorithm for further selection.

Main Results:

• Theoretical strong screening consistency is established for the proposed procedure.
• The quasi-maximum likelihood estimator for influential powers demonstrates asymptotic normality.
• The method effectively identifies portal nodes in large-scale social networks.
• Numerical studies on a Sina Weibo dataset validate the approach.

Conclusions:

• The developed NAM-based approach provides a theoretically sound and practically effective method for identifying influential users.
• The screening and estimation procedures offer reliable tools for analyzing network influence.
• The findings contribute to a deeper understanding of user influence in complex social networks.