Attributed graph clustering with multi-task embedding learning.

Area of Science:

• Data Science
• Machine Learning
• Graph Theory

Background:

• Attributed graph clustering requires integrating both graph structure and node features.
• Existing methods using graph neural networks often produce embeddings not optimized for clustering.
• Current loss functions may not fully retain both structural and feature information.

Purpose of the Study:

• To propose a novel multi-task embedding learning method (MTEL) for attributed graph clustering.
• To develop node embeddings that effectively capture both structural and feature information for improved clustering.
• To enhance the performance of attributed graph clustering by addressing limitations in current embedding techniques.

Main Methods:

• Developed a multi-task embedding learning framework (MTEL).
• Constructed two prediction tasks based on structure and feature adjacency matrices.
• Incorporated minimum hop number prediction to encode node correlation degrees into embeddings.
• Applied L2,1 norm regularization for joint learning of model parameters.

Main Results:

• MTEL demonstrated superior performance in attributed graph clustering compared to state-of-the-art methods.
• The proposed method effectively integrates graph structure and node features into learned embeddings.
• Experiments on real attributed graphs validated the efficacy of MTEL.

Conclusions:

• MTEL offers an effective approach for attributed graph clustering by optimizing node embeddings.
• The multi-task learning strategy enhances the retention of both structural and feature information.
• MTEL represents a significant advancement in attributed graph clustering techniques.