Temporal Dynamics Enhanced Sequential Recommendation with Adaptive Attention Mechanism

Sequential recommendation aims to predict future user actions by analyzing interactions between users and items. While most existing methods primarily focus on sequences of user behaviors, they tend to neglect the temporal dynamics between interactions, which can hinder the ability to capture evolving user interests. To address this limitation, we propose TDESRec, a model that embeds the time intervals between user behaviors through sinusoidal coding. By embedding temporal intervals, the model dynamically adjusts the significance of each behavior in the sequence, combining item similarity with time-aware attention to better capture temporal patterns. The performance of the model is evaluated on two datasets, demonstrating improvements in accuracy and robustness. The results highlight the effectiveness of incorporating temporal dynamics into sequential recommendation tasks and show that TDESRec enhances performance in terms of predictive accuracy and user interest adaptation.