Hop-Aware Linear Transformers for Scalable Graph Neural Networks

Node classification is a fundamental task in graph representation learning, with widespread applications in domains such as social networks, bioinformatics, and knowledge graphs. Existing methods, however, face significant challenges when applied to large-scale graphs due to high computational complexity and limited ability to capture longrange node dependencies. To address these issues, we propose HALT-GNN, an enhanced Graph Transformer model that efficiently models and integrates both local and global semantics. HALT-GNN encodes multi-hop neighborhood features into sequences and incorporates a global attention mechanism, enabling the model to capture complex graph structures while maintaining computational efficiency. Our model's modular design allows it to effectively handle large-scale graph data, achieving significant improvements in both prediction accuracy and computational efficiency. Experimental results across multiple benchmark datasets demonstrate the superior performance of HALT-GNN, providing a novel and efficient solution for large-scale graph representation learning.