Quantum-PSO based unsupervised clustering of users in social networks using attributes

Area of Science:

• Social Network Analysis
• Data Mining
• Machine Learning

Background:

• Unsupervised cluster detection groups similar users in social networks.
• Existing methods often use links or both links and attributes.
• Attribute-based clustering is valuable but K-mode can face local optima.

Purpose of the Study:

• To propose a novel clustering method using only user attributes.
• To address the limitations of the K-mode algorithm in social network analysis.
• To enhance the accuracy of social network user clustering.

Main Methods:

• Attribute-based dimensionality reduction (attribute selection and removal).
• Quantum Particle Swarm Optimization (QPSO) for similarity maximization.
• Utilized three distinct similarity measures for attribute reduction and clustering.

Main Results:

• The proposed Quantum PSO approach demonstrated superior clustering performance.
• Outperformed K-Mode and K-Mean algorithms on ego-Twitter and ego-Facebook datasets.
• Achieved better results across three key performance metrics.

Conclusions:

• The Quantum PSO method effectively detects social network clusters based on user attributes.
• This approach offers an improved alternative to traditional clustering algorithms for categorical data.
• The methodology enhances user similarity maximization for more accurate social network segmentation.