Temporal social network modeling of mobile connectivity data with graph neural networks

Area of Science:

• Complex networks analysis
• Machine learning applications
• Social network modeling

Background:

• Graph neural networks (GNNs) excel at modeling complex network connectivity.
• Temporal social network analysis using mobile connectivity data is underexplored.
• Predicting user activity in mobile networks is crucial for understanding social dynamics.

Purpose of the Study:

• To evaluate snapshot-based temporal GNNs for predicting mobile communication activity.
• To compare GNN performance against a non-GNN baseline (EdgeBank).
• To assess the potential of GNNs in temporal social network analysis.

Main Methods:

• Investigated four snapshot-based temporal GNN models.
• Developed a non-GNN baseline using the EdgeBank method.
• Analyzed phone call and SMS activity data from a mobile network.

Main Results:

• The ROLAND temporal GNN model demonstrated superior performance over the baseline in most scenarios.
• Three other temporal GNN models performed worse than the baseline on average.
• GNN-based approaches show potential for temporal social network analysis.

Conclusions:

• GNNs hold promise for analyzing temporal social networks via mobile connectivity data.
• The ROLAND GNN is a promising architecture, outperforming the baseline.
• Further research into specialized GNN architectures is necessary due to small performance margins.