Power Attributed Graph Embedding and Clustering.

Area of Science:

• Data Science
• Network Analysis
• Machine Learning

Background:

• Representation learning and node clustering are key tasks in attributed network analysis.
• Existing methods often perform these tasks sequentially, missing opportunities for synergistic improvement.
• Attributed networks combine node features with network structure, posing unique analytical challenges.

Purpose of the Study:

• To develop a novel method that jointly performs node embedding and clustering for attributed networks.
• To improve the accuracy and efficiency of representation learning and partitioning in attributed networks.
• To address the limitations of sequential processing in current attributed network analysis techniques.

Main Methods:

• Proposed the Power-Attributed Graph Embedding and Clustering (PAGEC) algorithm.
• Introduced a novel powered proximity matrix to jointly encode node link and attribute affinities.
• Formulated a new matrix decomposition model for simultaneous embedding and clustering.

Main Results:

• PAGEC effectively integrates node representation learning and clustering.
• The powered proximity matrix captures complex data affinities.
• Experimental results show superior performance compared to state-of-the-art methods, including deep learning approaches.
• Theoretical analysis confirms the link between the proximity matrix and random walk theory.

Conclusions:

• Jointly optimizing embedding and clustering in attributed networks yields significant performance gains.
• PAGEC offers a robust and effective approach for attributed network analysis.
• The method demonstrates broad applicability across diverse attributed network datasets.