Dual-Pathway Feature Fusion Graph Transformer

Graph Neural Networks (GNNs) still suffer from performance degradation in node classification tasks due to information aggregation limitations and the over-smoothing phenomenon. To address these challenges, this paper proposes an innovative GNN model named the Dual-Pathway Feature Fusion Graph Transformer (DPFFGT), which combines a global attention mechanism with node feature convolution.this work introduces a global attention mechanism from graph Transformers. By employing parallel weighted operations, this approach enhances the neighborhood information aggregation capability of Graph Convolutional Networks (GCNs) while preventing information loss. Experimental results demonstrate that the proposed model achieves significant performance improvements in node classification tasks. Particularly on imbalanced datasets, it effectively enhances classification accuracy, mitigates over-smoothing, and optimizes graph structures. This study provides a novel solution for applying graph neural networks to node classification tasks, offering both theoretical and practical value.