Efficient Recommendation with Dynamic Subgraph Enhanced GCN*

This study proposees DCSGCN (Dynamic Collaborative Subgraph enhanced Graph Convolutional Network), a novel Graph Neural Network (GNN)-based recommendation model that addresses the computational efficiency challenges in large-scale recommend-dation systems. The proposed approach integrates dynamic subgraph sampling strategies and a caching mechanism to reduce computational overhead while preserving recommendation quality. Experimental results on the MovieLens <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1M}$</tex> dataset demonstrate that DCSGCN achieves a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{6 7. 4 5 \%}$</tex> reduction in evaluation time compared to LightGCN, while maintaining competitive performance across standard recommendation metrics. These findings highlight DCSGCN particularly suitable for large-scale recommendation scenarios where computational efficiency is critical.