This study investigates the application of graph neural networks (GNN) in session-based recommendation systems (SR), focusing on a key modification involving the use of a global vector. Session-based recommendation systems often face challenges in accurately capturing user behavior due to the limited data available within individual sessions. The SR-GNN model, originally designed for automatic feature extraction from session graphs by leveraging rich connections between nodes, addresses these challenges effectively. In our experiments, we replaced the local vector with a global vector representing the entire session sequence, not just the last element. Our results show that both local and global vectors perform comparably, suggesting that the global vector is sufficient to capture the session context. Additionally, our study indicate that the SR-GNN algorithm maintains consistent performance across various datasets, with minor fluctuations depending on the dataset characteristics. The conducted experiments highlight the resilience and adaptability of the SR-GNN model in diverse scenarios, demonstrating its potential for use in session-based recommendation systems.
Authors
Additional information
- DOI
- Digital Object Identifier link open in new tab 10.1007/978-3-031-70421-5_22
- Category
- Publikacja monograficzna
- Type
- rozdział, artykuł w książce - dziele zbiorowym /podręczniku w języku o zasięgu międzynarodowym
- Language
- angielski
- Publication year
- 2024
