Heterogeneous graph convolutional network for multi-view semi-supervised classification.

Area of Science:

• Machine Learning
• Graph Neural Networks
• Data Mining

Background:

• Existing multi-view learning methods often process data views independently, neglecting inter-view semantic interactions.
• Recent graph convolutional network (GCN) approaches aggregate view-specific representations, failing to model complex relationships within heterogeneous data.

Purpose of the Study:

• To propose a unified framework for semantic representation learning from multi-view datasets by treating them as a heterogeneous graph.
• To address the limitations of existing methods in capturing cross-view semantic dependencies.

Main Methods:

• A novel approach models multi-view data as a heterogeneous graph with shared nodes and view-specific edge types.
• Utilizes a heterogeneous graph convolutional network (HGCN) to extract semantic representations.
• Employs an adaptively learned graph Laplacian matrix from multi-type edges.

Main Results:

• The proposed HGCN-MVSC method demonstrates superior performance in semi-supervised classification tasks.
• Achieved encouraging results across eight public datasets compared to state-of-the-art competitors.

Conclusions:

• Representing multi-view data as a heterogeneous graph is a promising direction for representation learning.
• The HGCN-MVSC framework effectively captures complex semantic interactions for enhanced semi-supervised classification.