Path-enhanced graph convolutional networks for node classification without features.

Area of Science:

• Graph Neural Networks
• Machine Learning
• Network Analysis

Background:

• Current graph neural networks (GNNs) often overlook inherent graph characteristics, potentially limiting performance.
• Few methods address the impact of these characteristics, especially in graphs lacking node features.

Purpose of the Study:

• To improve the performance of graph convolutional networks (GCNs) on graphs without node features.
• To introduce a novel approach that leverages graph structure for enhanced node classification.

Main Methods:

• Proposing t-hopGCN, a method that describes t-hop neighbors using shortest path distances.
• Utilizing the adjacency matrix of t-hop neighbors as features for node classification.
• Integrating t-hop neighbor information into existing popular GNN architectures.

Main Results:

• t-hopGCN significantly enhances node classification performance on graphs lacking node features.
• The inclusion of t-hop neighbor adjacency matrices improves the efficacy of established GNN models.
• Demonstrated superior performance compared to baseline methods in node classification tasks.

Conclusions:

• Leveraging t-hop neighbor information is crucial for improving GNNs, particularly in feature-less graph scenarios.
• The proposed t-hopGCN method offers a viable solution for enhancing node classification.
• The approach is generalizable and can benefit various existing GNN architectures.