I²HGNN: Iterative Interpretable HyperGraph Neural Network for semi-supervised classification.

Area of Science:

• Machine Learning
• Graph Theory
• Data Science

Background:

• Hypergraphs capture complex higher-order interactions beyond traditional graphs.
• Existing hypergraph neural networks often suffer from information distortion or lack theoretical grounding.

Purpose of the Study:

• To propose a novel hypergraph neural network, I²HGNN, addressing limitations of current methods.
• To establish a theoretically sound approach for hypergraph learning.

Main Methods:

• Developed I²HGNN based on an energy minimization function for hypergraphs.
• Aligned propagation layers with the message-passing paradigm for hypergraphs.
• Evaluated performance across 15 diverse datasets for node classification.

Main Results:

• I²HGNN demonstrates a favorable balance between performance and interpretability.
• The model effectively integrates node features and hypergraph topology.
• Achieved superior performance in hypergraph node classification on nearly all benchmark datasets.

Conclusions:

• I²HGNN offers a robust and effective solution for hypergraph learning.
• The energy minimization framework provides a solid theoretical foundation.
• The proposed method advances the field of hypergraph neural networks.