Link prediction plays a crucial role in graph analysis and is widely used in areas like social network analysis, bioinformatics, and recommendation systems. The majority of existing techniques depend on a solitary model, often making it challenging to accurately represent the graph’s overall structure and intricate nonlinear connections, thereby restricting the precision of predictions. In response to this issue, the paper presents an enhanced structure that amalgamates SVD with VGAE. SVD is proficient in capturing the global structure of the graph and can generate low-dimensional representations that retain basic global features. The VGAE learns in the latent space. This is used to model the nonlinear relationship between nodes. In this paper, these two techniques are placed in a unified framework with a joint loss function. Consequently, the model enhances the depiction of information on both a global and local scale. Experimental outcomes on various benchmark datasets indicate the superiority of this model over the standalone SVD and VGAE techniques.