Detecting chaotic behaviors in dynamic complex social networks using a feature diffusion-aware model.

Area of Science:

• Complex Systems Science
• Network Science
• Data Mining

Background:

• Dynamic complex social networks exhibit evolving structures and feature interactions.
• Detecting chaotic behaviors is crucial for understanding network dynamics and identifying anomalies.
• Existing methods may not fully capture the influence of feature diffusion on network anomalies.

Purpose of the Study:

• To propose a novel feature diffusion-aware model for detecting chaotic behaviors in dynamic complex social networks.
• To identify chaotic behaviors from the perspectives of abnormal links and abnormal nodes.
• To validate the model's effectiveness using real-world social network datasets.

Main Methods:

• Constructing a probabilistic model of dynamic complex social networks.
• Incorporating feature dynamics, node feature evolution, feature diffusion, and link generation.
• Utilizing Markov Chain Monte Carlo (MCMC) sampling methods (Metropolis-Hastings, Slice sampling) for parameter extraction.
• Measuring deviations from the model to detect chaotic behaviors.

Main Results:

• The proposed model demonstrated improved performance in detecting chaotic behaviors.
• Key performance metrics including accuracy, F1-score, Matthews Correlation Coefficient, recall, precision, AUC, and log-likelihood showed significant enhancements.
• Validation on Google+ and Twitter datasets confirmed the model's efficacy.

Conclusions:

• The feature diffusion-aware model effectively detects chaotic behaviors in dynamic social networks.
• The approach offers a robust method for identifying anomalies in complex network structures.
• This work provides a valuable contribution to the field of network anomaly detection.