A Multiobjective Evolutionary Algorithm Based on Structural and Attribute Similarities for Community Detection in...

Area of Science:

• Graph theory
• Network science
• Data mining

Background:

• Existing community detection algorithms primarily rely on vertex connectivity.
• Real-world networks often feature vertex attributes crucial for clustering.
• Simultaneously considering topological structure and vertex properties is a key challenge.

Purpose of the Study:

• To propose a multiobjective evolutionary algorithm (MOEA-SA) for attributed graph clustering.
• To integrate both structural and attribute similarities into community detection.
• To address the challenge of balancing topological and attribute information.

Main Methods:

• Developed a multiobjective evolutionary algorithm based on structural and attribute similarities (MOEA-SA).
• Introduced 'attribute similarity' as a novel objective alongside modularity.
• Employed a hybrid representation and a neighborhood correction strategy for gene repair.
• Designed a multi-individual-based mutation operator to guide evolutionary search.

Main Results:

• Validated MOEA-SA on real Facebook attributed graphs and ego-networks.
• Evaluated community quality using density and entropy metrics.
• MOEA-SA achieved superior density and entropy values compared to existing methods.
• Identified relevant communities with practical significance.

Conclusions:

• MOEA-SA effectively detects communities in attributed graphs by integrating structural and attribute information.
• The algorithm demonstrates superior performance over existing methods in identifying meaningful communities.
• Knee points provide decision support for selecting optimal community structures.