One of the greatest challenges of current recommender systems is still accurately predicting commonalities between users. To improve the prediction of comparable user profiles, this research proposes a new method that combines bipartite graphs with the structural similarity measure SimRank. Our solution captures both direct and indirect links between users by applying SimRank and modeling users and their interests in a bipartite graph. This method provides more precise and consistent predictions than traditional methods such as cosine similarity, according to experimental results obtained with the MovieLens dataset. The combination of SimRank and a bipartite structure offers a solid foundation for recognizing subtle and complex similarities between users. The present study demonstrates the effectiveness of the proposed approach and establishes a framework for future investigations, which may involve integrating contextual information and advanced techniques like Graph Neural Networks to further enhance recommender system performance.