<p>This research introduces a bilinear diffusion graph convolutional network (BiDGCN) to address the pervasive issue of data sparsity in collaborative filtering recommendation systems. By integrating user social interactions with graph convolutional networks, the BiDGCN captures both user–neighbor and neighbor–neighbor interactions through diffusion and bilinear aggregation mechanisms. This dual aggregation approach dynamically models social influence, significantly enhancing user representation and recommendation accuracy. Experimental evaluations on the Yelp and Flickr datasets demonstrate the efficacy of the model, achieving a 5% higher hit rate and a 6% improvement in normalized discounted cumulative gain compared to state-of-the-art models. The BiDGCN provides a robust solution for social recommendation tasks, overcoming data sparsity while delivering superior predictive performance.</p>