MECCH: Metapath Context Convolution-based Heterogeneous Graph Neural Networks.

Area of Science:

• Graph Representation Learning
• Artificial Intelligence
• Machine Learning

Background:

• Heterogeneous Graph Neural Networks (HGNNs) are used for representation learning on complex structural data.
• Deep HGNNs face performance degradation; metapath integration aims to improve semantic associations.
• Existing metapath-based models struggle with information loss and high computational costs.

Purpose of the Study:

• To address the limitations of existing metapath-based HGNNs.
• To introduce a novel model, MECCH, for efficient and lossless information aggregation.
• To improve prediction accuracy and computational efficiency in heterogeneous graph analysis.

Main Methods:

• Developed a novel Metapath Context Convolution-based Heterogeneous Graph Neural Network (MECCH).
• Introduced metapath contexts for lossless and non-redundant node information aggregation.
• Implemented three key components: metapath context construction, encoder, and convolutional fusion.

Main Results:

• MECCH demonstrated superior prediction accuracy on five real-world heterogeneous graph datasets.
• The model achieved improved computational efficiency compared to state-of-the-art baselines.
• Experiments covered node classification and link prediction tasks.

Conclusions:

• MECCH effectively overcomes information loss and high computational costs in metapath-based HGNNs.
• The proposed model offers a significant advancement in heterogeneous graph representation learning.
• MECCH provides a computationally efficient and accurate solution for complex graph data analysis.