Grafik sinir ağları yöntemleri ile öneri sistemi tasarımı

In recent years, the use of Graph Neural Networks (GNNs) in recommender systems has gained increasing attention, owing to their capacity to represent complex relationships between users and items through feature embeddings. However, most existing graph-based systems primarily rely on rating and review data, limiting their ability to capture structural and positional information within the graph, as well as the dynamic nature of user preferences. This study proposes an advanced end-to-end Graph Neural Network architecture (SPT-GNN) that integrates state-of-the-art feature engineering, embedding techniques, positional encoding, topological feature extraction, and transfer learning strategies to overcome these limitations. A pre-trained encoder trained on the IMDb dataset facilitates knowledge transfer, effectively bridging domain gaps and enhancing prediction accuracy. The proposed model effectively learns both local and global user-item interactions and captures temporal-dynamic user preferences. Extensive experimental analyses conducted on the Movielens dataset demonstrate that the SPT-GNN model significantly outperforms comparative baseline models (Collaborative Filtering, Content-based Filtering, Matrix Factorization, Vanilla GNN, BIVAECF, RecVAE), achieving notable improvements in metrics such as accuracy, precision, recall, and area under the curve (AUC). These findings highlight the power of integrating related features, sophisticated embedding techniques, and transfer learning to enhance recommendation accuracy, coverage, and user satisfaction. Additionally, this research contributes to the field of GNN-based recommender systems by offering a model architecture that effectively integrates structural and temporal information, providing a solid foundation for future research and practical applications.