Graph embedding based label propagation for community detection in social networks.

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Real-world networks exhibit community structures crucial for understanding their organization.
• Traditional community detection methods, like the Label Propagation Algorithm (LPA), are computationally efficient but often lack accuracy.
• LPA's limitations stem from its naive neighbor selection during label propagation, impacting modularity and structural integrity.

Purpose of the Study:

• To develop an improved community detection algorithm that addresses the limitations of conventional LPA.
• To enhance the accuracy and structural understanding of community detection in complex networks.
• To create a hybrid approach combining local connectivity with global structural information.

Main Methods:

• Developed Embedding-based Label Propagation (ELP), a hybrid algorithm integrating LPA with node embedding techniques.
• ELP's update step considers both local neighborhood information and embedding-based similarities for neighbor selection.
• Evaluated ELP on benchmark datasets (Karate Club, Dolphins, Football, Polbooks) and LFR synthetic networks, comparing performance against LPA and other established algorithms.

Main Results:

• ELP consistently outperformed LPA and other algorithms in terms of modularity, Normalized Mutual Information (NMI), and F1-score (NF1).
• The algorithm demonstrated scalability, performing effectively on large and complex networks.
• Empirical findings confirm ELP's superior performance in identifying accurate community structures.

Conclusions:

• Embedding-based Label Propagation (ELP) offers a significant advancement in network community detection.
• ELP provides a powerful and effective method for uncovering community structures in both real-world and synthetic networks.
• The integration of node embeddings enhances the accuracy and robustness of community detection algorithms.