Link Prediction in Dynamic Social Networks Combining Entropy, Causality, and a Graph Convolutional Network Model.

Area of Science:

• Social Network Analysis
• Data Mining
• Machine Learning

Background:

• Dynamic social networks exhibit complex topology and temporal evolution, posing challenges for link prediction.
• Understanding social relationship evolution is crucial for analyzing network dynamics.

Purpose of the Study:

• To propose an innovative fusion framework for link prediction in dynamic social networks.
• To enhance the accuracy and effectiveness of predicting future connections in evolving networks.

Main Methods:

• Preprocessing raw data to extract timestamp information.
• Introducing Temporal Information Entropy (TIE) integrated with Node2Vec for initial node feature generation.
• Applying causality analysis for secondary feature processing.
• Constructing an equal dataset by adjusting positive and negative sample ratios.
• Training a dedicated Graph Convolutional Network (GCN) model.

Main Results:

• The proposed framework demonstrated superior performance compared to existing methods.
• Key evaluation indicators including precision, recall, F1 score, and accuracy were significantly improved.
• Extensive experiments on multiple real-world social networks validated the framework's effectiveness.

Conclusions:

• The fusion framework offers a fresh perspective on predicting link dynamics in social networks.
• The study highlights the practical value of integrating entropy, causality, and GCN for robust link prediction.
• This research contributes to a deeper understanding of social relationship evolution in dynamic networks.