Multi-view heterogeneous graph learning with compressed hypergraph neural networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Multi-view learning aims to improve prediction by integrating multiple feature types from a single instance.
• Existing graph-based methods often rely on homogeneous assumptions and pairwise relationships, limiting their ability to capture complex inter-instance interactions.
• Real-world data frequently exhibits heterogeneous features and higher-order correlations that are not adequately addressed by current approaches.

Purpose of the Study:

• To develop a novel graph-based multi-view learning framework that addresses the limitations of existing methods.
• To effectively capture rich multi-view heterogeneous semantic information and higher-order correlations.
• To improve compatibility prediction in multi-view scenarios.

Main Methods:

• Designed a compressed hypergraph neural network tailored for multi-view heterogeneous graph learning.
• Incorporated a hypergraph structure to explore higher-order correlations among samples.
• Utilized efficient hypergraph convolutional networks within an explainable regularizer-centered optimization framework.
• Applied low-rank approximation to reformat complex initial multi-view heterogeneous graphs.

Main Results:

• The proposed compressed hypergraph neural network effectively captures multi-view heterogeneous semantic information.
• The hypergraph structure successfully models higher-order correlations between samples in multi-view settings.
• Extensive experiments demonstrated the method's feasibility and effectiveness compared to advanced node and multi-view classification techniques.

Conclusions:

• The developed compressed hypergraph neural network offers a powerful approach for multi-view heterogeneous graph learning.
• This method enhances compatibility prediction by better modeling complex data interactions.
• The findings suggest a promising direction for future research in multi-modal fusion and graph representation learning.