Iterative Deep Structural Graph Contrast Clustering for Multiview Raw Data.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Multiview clustering aims to group data instances without manual labels.
• Traditional methods rely on raw features, while deep methods often neglect data structure.
• Existing approaches have limitations in leveraging both feature quality and structural information.

Purpose of the Study:

• To propose a novel iterative deep structural graph contrast clustering (IDSGCC) method for multiview raw data.
• To address limitations of traditional and deep multiview clustering by integrating topology and representation learning.
• To enhance clustering performance by preserving and utilizing data structure information.

Main Methods:

• Developed a method combining topology learning (TL), representation learning (RL), and graph structure contrastive learning.
• TL module obtains a structured global graph to guide RL.
• RL module uses graph convolutional networks (GCN) with structural graphs and raw features.
• Graph structure contrastive learning operates on the similarity matrix, not just sample representations.
• An iterative update mechanism refines the data topology for improved clustering.

Main Results:

• The proposed IDSGCC method achieves improved clustering-friendly embeddings.
• Iterative topology updates lead to more credible data structures and better clustering.
• Experimental results on eight multiview datasets demonstrate superior performance over state-of-the-art methods.
• The model effectively integrates structural information with deep learning for enhanced clustering.

Conclusions:

• IDSGCC offers a robust framework for multiview clustering by synergistically combining topology, representation, and contrastive learning.
• The iterative mechanism is crucial for refining data topology and achieving high clustering accuracy.
• This approach advances the field of deep multiview clustering by effectively leveraging structural information.