Joint Latent Space Model for Social Networks with Multivariate Attributes.

Area of Science:

• Social, behavioral, and economic sciences
• Network analysis
• Data science

Background:

• Modeling social networks and high-dimensional attributes is crucial in social sciences.
• Existing methods may not effectively integrate network structure with individual characteristics.

Purpose of the Study:

• To propose a joint latent space model (JLSM) for integrating social network data and multivariate attributes.
• To develop a robust estimation algorithm for attribute and person locations within this joint space.
• To enable effective visualization and prediction of social networks and attributes.

Main Methods:

• Developed a joint latent space model (JLSM).
• Implemented a variational Bayesian expectation-maximization algorithm for parameter estimation.
• Applied the JLSM to analyze the social networks and attributes of French financial elites.
• Utilized the JLSM for analyzing user networks and behaviors on multimodal social media platforms like YouTube.

Main Results:

• The JLSM effectively summarizes information from social networks and multivariate attributes.
• A clear division within the social circles of French elites was observed, correlating with attribute differences.
• The methodology facilitates informative visualization and prediction capabilities for network and attribute data.

Conclusions:

• The JLSM provides a powerful framework for joint analysis of social networks and individual attributes.
• The model offers insights into social structures and attribute-driven divisions within groups.
• An R package 'jlsm' is available for practical application of the proposed methodology.