Classifying Dissemination Processes in Temporal Graphs.

Area of Science:

• Computer Science
• Network Science
• Data Mining

Background:

• Real-world graphs are often temporal, capturing interactions over time.
• Temporal information is crucial for understanding dissemination processes (e.g., rumors, diseases).
• Existing static graph classification methods fail to leverage this temporal data.

Purpose of the Study:

• To develop a framework for temporal graph classification.
• To adapt existing graph kernels and neural networks for temporal data.
• To improve classification accuracy for dynamic network processes.

Main Methods:

• Proposed a framework to extend standard graph kernels and neural networks to the temporal domain.
• Explored three distinct temporal approaches, analyzing trade-offs between information loss and efficiency.
• Introduced stochastic kernel variants for scalability on large graphs, with approximation guarantees.

Main Results:

• Temporal methods significantly outperformed static approaches in classification accuracy on real-world social networks.
• The framework demonstrated scalability for large graphs and datasets.
• High classification accuracy was achieved even with incomplete dissemination process information.

Conclusions:

• The proposed temporal graph framework effectively captures dynamic processes, outperforming static methods.
• The methods are both accurate and scalable, suitable for large-scale temporal network analysis.
• The framework shows promise for applications with partial data on information diffusion.