Enhancing Session-Based Recommendations with Popularity-Aware Graph Neural Networks
Qingbo Sun, Weihua Yuan, Qi Zhang, Zhijun Zhang
Abstract
Real-time and reliable recommendations are essential for anonymous users in session-based recommendation systems. Graph neural network-based algorithms are attracting more researchers due to their simplicity and efficiency. However, current methods overlook the influence of edge frequency on feature aggregation in graph modeling and fail to account for the impact of item popularity on user interest. To address these issues, a novel approach called Popularity-Aware Graph Neural Networks for Session-based Recommendations is proposed. This study integrates both edge frequency and item popularity into the modeling process to enhance the learning of item features and user interests. A graph that includes the number of edge occurrences is constructed, and a graph neural network with an attention mechanism is utilized to learn user interests and item features by aggregating information from the graph. Finally, the session's final representation is learned based on the occurrence frequency of items. The proposed study evaluates the model on two classical ecommerce datasets and demonstrates its superiority over existing methods.