Dynamic Graph Learning for Systems Using Selective State Focused Attention Networks
Shikhar Vashistha, Neetesh Kumar · 2025
Traditional graph neural networks (GNNs) lack scalability and lose individual node characteristics due to oversmoothing, especially in the case of deeper networks. This results in sub-optimal feature representation, affecting the model's performance on tasks involving dynamically changing graphs. To address this issue, we present Graph Selective state focused Attention Networks (GSANs) based neural network architecture for graph-structured data. The GSAN incorporates multi-head masked self-attention (MHMSA) and sequential state space modeling (S3M) layers to overcome the limitations of traditional GNNs. In GSAN, the MHMSA allows GSAN to dynamically emphasize crucial node connections, particularly in evolving graph environments. The S3M layer enables the network to track and adjust dynamically in changing node states and improving predictions of node behavior in varying contexts. The S3M layer enhances the generalization of unseen graph structures and provides interpretability by analyzing how node states affect the relative importance of links within the graph. With this, GSAN effectively outperforms inductive and transductive tasks and overcomes the issues that traditional GNNs experience. To analyze the performance behavior of GSAN, a set of state-of-theart (SOTA) comparative experiments are conducted on graphs benchmark datasets, including Cora, Citeseer, Pubmed network citation, and protein-protein-interaction datasets. As an outcome, GSAN improved the classification accuracy by 0.833%, 0.5%$, 0.37%, and 1.54 % on the F1-score, respectively. The code is publicly available at https://github.com/shikharvashistha/GSAN.