Visual Analytics of Multivariate Networks With Representation Learning and Composite Variable Construction

Area of Science:

• Data Science
• Network Analysis
• Information Visualization

Background:

• Multivariate networks are prevalent in real-world data applications.
• Understanding relationships within these networks is challenging.
• Existing methods lack intuitive interpretation of complex network characteristics.

Purpose of the Study:

• To present a visual analytics workflow for studying multivariate networks.
• To extract associations between structural and semantic network characteristics.
• To simplify complex network data for user interpretation.

Main Methods:

• A neural network-based learning phase for data classification.
• A dimensionality reduction and optimization phase for result simplification.
• An interactive visualization interface for user interpretation.
• Composite variable construction to linearize nonlinear features.

Main Results:

• Demonstrated workflow capabilities on social media network data.
• Successfully extracted associations between network attributes.
• Enabled intuitive interpretation of complex network features through visualization.
• Validated through expert qualitative feedback.

Conclusions:

• The proposed visual analytics workflow effectively aids in understanding multivariate networks.
• The workflow facilitates the discovery of associations between network properties.
• Composite variable construction enhances the interpretability of neural network outputs for network analysis.