Community evolution prediction based on feature change patterns in social networks.

Area of Science:

• Social Network Analysis
• Machine Learning
• Data Mining

Background:

• Predicting community evolution in dynamic social networks is vital for trend analysis and proactive safety measures.
• Existing methods struggle with feature extraction in highly interactive networks, limiting predictive accuracy.

Purpose of the Study:

• To propose a novel community evolution prediction method based on feature change patterns.
• To develop an algorithm for learning feature change rules and capturing dynamic community information.
• To enhance prediction accuracy and efficiency in dynamic social networks.

Main Methods:

• Developed a community evolution prediction method focusing on feature change patterns.
• Designed an algorithm to learn feature change rules and identify community feature patterns.
• Implemented a parallel learning strategy with parameter sharing for efficiency.

Main Results:

• The proposed method achieved approximately 25% improvement in predictive performance across multiple datasets (AS, DBLP, Facebook).
• The feature change pattern approach captured richer dynamic information than static state features.
• The parallel learning mechanism reduced training time by nearly half compared to baseline methods.

Conclusions:

• Community evolution prediction based on feature change patterns is more effective than traditional state-feature methods.
• The proposed method provides a robust framework for understanding and forecasting community dynamics.
• The parallel learning strategy enhances computational efficiency for large-scale network analysis.