Popularity-Aware Graph Neural Network with Global Context for Session-Based Recommendation

Xiangwei Zeng, Chao Chang, Feiyi Tang, Zhengyang Wu, Yong Tang

Abstract: Session-based recommendation aims to predict the next interaction in an anonymous user’s sequence and has gained significant attention. Most existing systems model user preferences from the current session using graph neural networks but overlook the varying importance of items with different popularity. To address this, we propose the Popularity-aware Graph Neural Network with Global Context (PGNN-GC), which models popularity features to better capture users’ diverse preferences. By explicitly modeling popularity-aware embeddings and using attention mechanisms, PGNN-GC differentiates user preferences for items of varying popularity. Additionally, we enhance representations using a contrastive learning paradigm. Experiments on three open datasets show that PGNN-GC achieves state-of-the-art performance.