Scalable and effective negative sample generation for hyperedge prediction.

Area of Science:

• Complex Systems Analysis
• Network Science
• Machine Learning

Background:

• Hypergraphs excel at modeling complex systems by capturing multi-entity interactions, outperforming traditional graphs.
• Hyperedge prediction is crucial for analyzing hypergraphs, requiring effective negative hyperedge sampling for model training.
• Existing negative sampling methods lack generalizability and scalability, especially for large hypergraphs.

Purpose of the Study:

• To develop a scalable and effective method for generating informative negative hyperedges for hyperedge prediction.
• To adapt diffusion models for the discrete space of negative hyperedge generation in hypergraphs.
• To improve the accuracy and scalability of hyperedge prediction models.

Main Methods:

• Introduced SEHP (Scalable and Effective Negative Sample Generation for Hyperedge Prediction), a conditional diffusion model for iterative negative hyperedge generation and refinement.
• Developed sub-hypergraph sampling techniques to integrate global structural information for scalable batch training.
• Addressed the challenges of applying diffusion models to discrete negative sample generation.

Main Results:

• SEHP effectively generates and refines negative hyperedges, pushing them towards the decision boundary to enhance model performance.
• The method demonstrates superior scalability by enabling batch training on sampled sub-hypergraphs.
• Extensive experiments on real-world datasets show SEHP outperforming state-of-the-art methods in prediction accuracy and scalability.

Conclusions:

• SEHP offers a significant advancement in negative hyperedge generation for hypergraph analysis.
• The proposed conditional diffusion model approach is effective and scalable for large hypergraphs.
• SEHP improves hyperedge prediction performance and addresses limitations of existing methods.