Movie Recommendation System Based on Lightweight Graph Convolutional Network

Recommender systems play a pivotal role in addressing information overload in digital environments. Traditional methods, constrained by linear assumptions and Euclidean vector space representations, often struggle with sparsity and fail to capture complex user–item relationships. Here, we present a lightweight movie recommendation framework based on Graph Convolutional Networks (GCNs), leveraging the MovieLens-32M dataset. By constructing a user–item bipartite graph and employing lightweight graph convolutional network for embedding learning, our model effectively captures high-order collaborative signals while maintaining computational efficiency. We formulate recommendation as a link prediction task, trained with Bayesian personalized ranking loss. Empirical evaluation demonstrates high precision (70.4%) and recall (22.3%) in Top-K recommendation, validating the model’s efficacy and scalability. This work highlights the potential of lightweight GCNs in large-scale, graph-based recommendation tasks and provides a foundation for further integration with dynamic and multimodal systems.