Coupled Node Similarity Learning for Community Detection in Attributed Networks.

Area of Science:

• Network Science
• Data Mining
• Machine Learning

Background:

• Attributed networks integrate network structure with node attributes.
• Existing community detection methods often overlook node attributes or complex attribute-structure interactions.
• Hierarchical couplings between attributes, nodes, and network structure are crucial for community formation.

Purpose of the Study:

• To introduce a novel coupled node similarity (CNS) method for attributed networks.
• To incorporate and learn complex attribute and structure couplings for node similarity computation.
• To improve community detection by considering hierarchical interactions.

Main Methods:

• Developed coupled node similarity (CNS) to compute similarity within and between nodes with categorical attributes.
• CNS integrates frequency-based intra-attribute similarity, co-occurrence-based inter-attribute similarity, and attribute-to-structure similarity.
• Transformed plain graphs into weighted graphs using CNS edge weights for clustering algorithms.

Main Results:

• CNS effectively captures attribute and structure couplings, including hierarchical interactions.
• CNS-based community detection algorithms demonstrated superior performance on various datasets compared to state-of-the-art methods.
• The approach successfully identified topologically well-connected and semantically coherent communities.

Conclusions:

• Coupled node similarity (CNS) offers a robust framework for community detection in attributed networks.
• Considering hierarchical attribute-structure interactions significantly enhances community detection accuracy.
• The proposed method provides a valuable tool for analyzing complex network data.