MAGNET: Multi-Interest Attentive Group Recommender with Deep Reinforcement Learning
Yaqi Shi, Xiaoqiang Ren, Xiaofan Wang · 2023
In real-life group activities, providing relevant content recommendations is challenging due to the dynamic nature of group interests. We propose MAGNET, a deep reinforcement learning-based group recommendation system. MAGNET employs multiple vectors to represent group interests collectively, addressing the challenge of inconsistent long-term and short-term preferences. We treat the recommendation task as a sequential decision problem, extracting points of interest from short-term interactions and using them to predict the next item. A neural network assigns weights to group members based on their current interests for a more accurate representation of long-term preferences. Our experiments on two datasets demonstrate the competitive performance of MAGNET, confirming its effectiveness. Ablation experiments highlight the contributions of individual components. MAGNET has the potential to enhance content recommendations for groups in real-life scenarios.