Booklizer - Um Aplicativo de Recomendação Literária Personalizada Usando GNN

Book recommendation systems like Goodreads consider user ratings and curated lists but still face limitations in terms of personalization, accuracy, and explainability. In this context, this project proposes the development of the Booklizer app, which combines a modern interface with a recommendation model based on Graph Neural Networks (GNNs), capable of capturing semantic relationships between books through their genres. The methodology was structured around: requirement definition; collection and cleaning of a literary dataset from Kaggle; prototyping with Figma; interface implementation using Flutter; and the construction of a bipartite graph between books and genres, on which a GNN was trained. The generated embeddings were evaluated using cosine similarity, revealing coherent thematic groupings and outlier detection. The partial results demonstrate the model’s potential to deliver satisfying recommendations aligned with readers’ profiles.