Deep Time-Aware Item Evolution Network for Sequential Recommendation

With the rapid advancement of information technology, recommendation systems have become essential tools for addressing information overload.While traditional methods such as content-based and collaborative filtering infer long-term preferences by modeling user-item interactions statically, they often fail to capture the dynamic and context-dependent nature of user preferences.Sequential recommendation methods address this limitation by modeling interaction histories as ordered sequences, thereby capturing item dependencies and short-term preferences.However, most existing models tend to overemphasize interaction order while neglecting temporal information, typically assuming uniform time intervals between interactions, which restricts their ability to capture the temporal dynamics of user preferences.To overcome these challenges, we propose the Time-Aware Item Evolution Network for Sequential Recommendation (DTNSRec), a novel model that integrates temporal information into an enhanced self-attention architecture.DTNSRec explicitly models the influence of dynamic time intervals on predicting the next item in a sequence.Additionally, it independently calculates positional correlations to mitigate noise and enhance the recognition of sequential patterns in user behavior.Extensive experiments conducted on four real-world datasets demonstrate that DTNSRec consistently outperforms a range of state-of-the-art sequential recommendation models, validating its effectiveness in capturing temporal dynamics and improving recommendation accuracy.