muxGNN: Multiplex Graph Neural Network for Heterogeneous Graphs.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Network Science

Background:

• Graph neural networks (GNNs) excel at network mining but are limited to homogeneous networks.
• Heterogeneous graphs, with diverse node and edge types, present unique modeling challenges.

Purpose of the Study:

• To introduce muxGNN, a multiplex graph neural network tailored for heterogeneous graphs.
• To develop a novel coupling attention mechanism for modeling multi-relational contexts.

Main Methods:

• Representing heterogeneous graphs as multiplex networks with relation layer graphs and a coupling graph.
• Parameterizing relation-specific node representations.
• Designing node invariant and equivariant coupling structures for different tasks.

Main Results:

• MuxGNN demonstrates superior performance on six real-world datasets for link prediction and graph classification.
• Experiments cover both transductive and inductive learning contexts.
• The coupling attention mechanism reveals interpretable cross-relational connections.

Conclusions:

• MuxGNN offers a powerful framework for analyzing heterogeneous graphs.
• The model advances the state-of-the-art in heterogeneous graph mining.
• Interpretable attention mechanisms enhance understanding of complex network structures.