Robust self supervised symmetric nonnegative matrix factorization to the graph clustering.

Area of Science:

• Network analysis
• Machine learning
• Data mining

Background:

• Graph clustering is crucial for network analysis, identifying node groups via similarities.
• Traditional Nonnegative Matrix Factorization (NMF) methods face challenges with noise, outliers, and nonlinear structures in real-world networks.

Purpose of the Study:

• To introduce Robust Self-Supervised Symmetric NMF (R3SNMF) for improved graph clustering.
• To enhance the accuracy and robustness of graph clustering algorithms.

Main Methods:

• R3SNMF employs a robust principal component model to mitigate noise and outliers.
• A self-supervised learning mechanism iteratively refines clustering and representations.
• Symmetric factorization preserves network structure, and graph-boosting enhances relationship representation.

Main Results:

• R3SNMF demonstrates superior performance compared to state-of-the-art methods.
• The algorithm shows enhanced accuracy and robustness on diverse real-world graph datasets.

Conclusions:

• R3SNMF effectively addresses limitations of traditional NMF in graph clustering.
• The proposed method offers a resilient and accurate approach for complex network analysis.