Community-aware biased random walks for community detection in attribute networks

Area of Science:

• Network analysis
• Data mining
• Machine learning

Background:

• Community detection is crucial for understanding network structures.
• Existing network embedding methods often ignore node attributes and require manual parameter tuning.
• Random walk-based methods face challenges with node attributes and parameter settings.

Purpose of the Study:

• To propose a community-aware biased random walk method for community detection in attributed networks.
• To address limitations of existing methods by incorporating node attributes and adaptive parameter settings.
• To improve the accuracy and efficiency of community detection.

Main Methods:

• A novel network embedding method based on community-aware biased random walk.
• Jointly leveraging topological and attribute similarity to guide random walks.
• Adaptive setting of random walk parameters based on node degrees and attributes.
• Utilizing the skip-gram model for learning node embeddings.
• Employing clustering algorithms for final community detection.

Main Results:

• The proposed method effectively identifies communities in attributed networks.
• It outperforms existing methods by considering both topological and attribute information.
• Adaptive parameter setting prevents oversampling of low-degree or low-attribute nodes.
• Experimental validation on real-world and synthetic datasets confirms effectiveness.

Conclusions:

• The community-aware biased random walk method offers a robust approach to community detection in attributed networks.
• Integrating node attributes and adaptive random walks enhances community identification accuracy.
• This method provides a more efficient and effective solution compared to traditional techniques.