Dual Information Enhanced Multiview Attributed Graph Clustering.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Multiview attributed graph clustering partitions data using attribute and adjacency information from multiple views.
• Graph Neural Networks (GNNs) show promise but often neglect view-specific information and struggle with representation recovery.
• Existing methods limit downstream clustering by failing to bridge low-level and high-level data representations.

Purpose of the Study:

• To propose a novel Dual Information Enhanced Multiview Attributed Graph Clustering (DIAGC) method.
• To address limitations in current GNN-based clustering by capturing both consensus and specific information.
• To enhance clustering performance through improved representation learning.

Main Methods:

• Introduced a Specific Information Reconstruction (SIR) module to disentangle consensus and specific information.
• Employed Contrastive Learning (CL) to align latent high-level and low-level representations.
• Integrated Self-Supervised Clustering (SC) to guide the high-level representation towards a desired clustering structure.

Main Results:

• The SIR module enables Graph Convolutional Networks (GCNs) to capture more essential low-level representations.
• The CL and SC modules facilitate the recovery of high-level representations suitable for clustering.
• DIAGC demonstrated superior performance compared to state-of-the-art methods on real-world benchmarks.

Conclusions:

• DIAGC effectively addresses the limitations of existing multiview graph clustering methods.
• The proposed method enhances clustering by leveraging both consensus and specific information across views.
• DIAGC offers a robust approach for partitioning complex multiview attributed graph data.