Learning the mechanisms of network growth

Area of Science:

• Network Science
• Computer Science
• Data Science

Background:

• Dynamic networks are complex systems where connections evolve over time.
• Selecting appropriate models for these networks is crucial for understanding their behavior.
• Existing methods struggle to accurately identify underlying network models.

Purpose of the Study:

• To propose a novel and highly accurate model-selection method for dynamic networks.
• To introduce innovative dynamic features for improved network classification.
• To validate the method on both synthetic and real-world network data.

Main Methods:

• Training a classifier on extensive synthetic network data generated from nine state-of-the-art dynamic random graph models.
• Developing novel dynamic features based on link reception counts for vertex groups within time intervals.
• Evaluating classification performance against existing state-of-the-art methods.

Main Results:

• The proposed method achieved near-perfect classification accuracy on synthetic dynamic networks.
• The new dynamic features proved to be computationally efficient, analytically tractable, and interpretable.
• The approach significantly outperformed current state-of-the-art model selection techniques.

Conclusions:

• The novel classification method offers a superior approach to model selection for dynamic networks.
• The method's application to citation networks supports existing literature on preferential attachment, fitness, and aging models.
• The findings highlight the effectiveness of the proposed dynamic features in characterizing network evolution.