Recommendation System for books using Graph Neural Networks

This paper analyses how GNNs is, GraphSAGE specifically can help develop a trustworthy book recommender system. It focuses on user-book interactions as a bipartite graph where users, and books are nodes connected via edges that represent these interactions or ratings. Using the Book-Crossing Dataset, where metadata are available, like ISBN, title, author, year, and also user demographics such as age, GraphSAGE model effectively aggregates information from the neighbouring nodes in order to learn embeddings which capture the user preferences, book attributes, and patterns of interactions. This is superior over traditional methods of collaborative filtering and matrix factorization because it deals with sparsity issues. Evaluation results (RMSE: 0.4818, MSE: 0.2321, MAE: 0.2872) show that the model is able to make accurate recommendations. This is further supported by visualizations of training loss and ratings distribution. This work shows the potential of GNNs, especially GraphSAGE, in improving recommendation systems. Future directions include scalability, time dynamics, and multi-modal data integration for enhanced personalization.