Local and High-Order Consistency Coding and Adaptation for Cross-Hypergraph Node Classification

Area of Science:

• Hypergraph learning
• Machine learning
• Data science

Background:

• Node classification is crucial in hypergraph learning, but acquiring labeled data is difficult in new hypergraphs.
• Existing methods often overlook high-order relationships, limiting representation discrimination.

Purpose of the Study:

• To address challenges in cross-hypergraph node classification by leveraging knowledge from a well-labeled source hypergraph to a target hypergraph.
• To develop a model that learns both discriminative and transferable node representations.

Main Methods:

• Proposes Local and High-order Consistency Coding and Adaptation (LHCCA) model.
• Exploits local and high-order consistency relationships within each hypergraph.
• Employs attention mechanisms for unified representations and adversarial domain adaptation with contrastive learning for feature transfer.

Main Results:

• LHCCA learns discriminative and transferable node representations.
• The model effectively utilizes both local and high-order consistency information.
• Extensive experiments show the proposed model's effectiveness on real-world datasets.

Conclusions:

• The LHCCA model offers an effective solution for cross-hypergraph node classification.
• It successfully addresses limitations of existing methods by incorporating high-order information and enabling knowledge transfer.
• Theoretical analyses support the model's desirable properties and practical performance.