Temporally Dynamic Session-Keyword Aware Sequential Recommendation System

Addressing the dynamic preferences and needs of users to provide highly personalized recommendations is a fundamental challenge in recommender systems. To tackle this challenge effectively, understanding both session and keyword information takes on critical significance. Despite the pivotal roles that these two elements play in user interactions, prior research has often approached them in isolation, without a concerted effort to jointly investigate their synergistic potential. To bridge this gap, we propose SeKeBERT4Rec, a novel recommendation model that leverages both session and keyword information within a transformer-based sequential framework. In doing so, we also fill the void between user preferences expressed through keywords and their dynamic behavioral patterns within sessions. Our contributions include introducing a holistic approach to recommendation by seamlessly integrating session and keyword data, conducting an extensive comparative analysis against state-of-the-art methods, and offering in-depth insights through an ablation study that underscores the individual contributions of each model component.