Hypergraph Semi-Supervised Contrastive Learning for Hyperedge Prediction Based on Enhanced Attention Aggregator.

Area of Science:

• Complex Systems Science
• Network Science
• Machine Learning

Background:

• Hyperedge prediction is vital for understanding complex systems but is hindered by node influence variations, hyperedge ordering, and sparse data.
• Existing methods struggle to capture higher-order relationships effectively due to these inherent challenges.

Purpose of the Study:

• To propose a novel framework, Order propagation Fusion Self-supervised learning for Hyperedge prediction (OFSH), to address limitations in current hyperedge prediction techniques.
• To enhance the accuracy and robustness of hyperedge prediction by modeling higher-order interactions and mitigating data sparsity.

Main Methods:

• OFSH utilizes a hyperedge order propagation mechanism with dynamic node importance weighting and max-min pooling for feature amplification.
• A key node-guided augmentation strategy with adaptive masking is employed to combat data sparsity and preserve semantic information.
• A triadic contrastive loss function is implemented to maximize cross-view consistency and capture invariant semantic features.

Main Results:

• OFSH demonstrated significant improvements in hyperedge prediction accuracy across five real-world hypergraph datasets.
• The proposed method outperformed existing state-of-the-art approaches in terms of Area Under the Receiver Operating Characteristic Curve (AUROC) and Average Precision (AP).

Conclusions:

• OFSH effectively addresses critical challenges in hyperedge prediction, including node heterogeneity, hyperedge order effects, and data sparsity.
• The framework offers a robust and accurate solution for uncovering higher-order relationships in complex systems, advancing the field of network science.