Hypergraph Collaborative Network on Vertices and Hyperedges.

Area of Science:

• Machine Learning
• Data Mining
• Graph Theory

Background:

• Real-world datasets often exhibit complex, non-pair-wise relationships.
• Hypergraphs offer a natural framework for modeling these intricate correlations.
• Existing deep learning models for hypergraphs primarily use vertex or hyperedge information separately.

Purpose of the Study:

• To propose a novel model, Hypergraph Collaborative Network (HCoN), for enhanced hypergraph representation learning.
• To leverage both vertex and hyperedge information synergistically for improved latent representations.
• To introduce a hypergraph reconstruction error for effective classifier learning.

Main Methods:

• Developed the Hypergraph Collaborative Network (HCoN) model.
• Integrated information from both vertices and hyperedges in a collaborative manner.
• Employed hypergraph reconstruction error as a regularizer for classification.

Main Results:

• HCoN achieves superior performance in semi-supervised vertex and hyperedge classification tasks.
• Experimental results on benchmark datasets show HCoN outperforms existing state-of-the-art methods.
• The model effectively learns informative latent representations by considering collaborative information.

Conclusions:

• The proposed HCoN model offers a significant advancement in hypergraph learning.
• Collaborative utilization of vertex and hyperedge information enhances classification accuracy.
• HCoN provides a robust framework for analyzing complex relational data.