![[2602.18283] HyTRec: A Hybrid Temporal-Aware Attention Architecture for Long Behavior Sequential Recommendation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18283] HyTRec: A Hybrid Temporal-Aware Attention Architecture for Long Behavior Sequential Recommendation
Summary
HyTRec introduces a Hybrid Temporal-Aware Attention architecture designed to enhance long behavior sequential recommendations, improving retrieval precision and efficiency.
Why It Matters
As recommendation systems increasingly rely on long sequences of user behavior, HyTRec addresses the trade-off between computational efficiency and retrieval accuracy. This innovation is crucial for industries that depend on precise recommendations from extensive user interaction data, making it a significant advancement in the field of information retrieval and AI.
Key Takeaways
- HyTRec combines linear and softmax attention mechanisms to balance efficiency and precision.
- The model effectively separates long-term preferences from short-term intents, enhancing recommendation accuracy.
- Empirical results show over 8% improvement in Hit Rate for users with ultra-long interaction sequences.
Computer Science > Information Retrieval arXiv:2602.18283 (cs) [Submitted on 20 Feb 2026] Title:HyTRec: A Hybrid Temporal-Aware Attention Architecture for Long Behavior Sequential Recommendation Authors:Lei Xin, Yuhao Zheng, Ke Cheng, Changjiang Jiang, Zifan Zhang, Fanhu Zeng View a PDF of the paper titled HyTRec: A Hybrid Temporal-Aware Attention Architecture for Long Behavior Sequential Recommendation, by Lei Xin and 5 other authors View PDF HTML (experimental) Abstract:Modeling long sequences of user behaviors has emerged as a critical frontier in generative recommendation. However, existing solutions face a dilemma: linear attention mechanisms achieve efficiency at the cost of retrieval precision due to limited state capacity, while softmax attention suffers from prohibitive computational overhead. To address this challenge, we propose HyTRec, a model featuring a Hybrid Attention architecture that explicitly decouples long-term stable preferences from short-term intent spikes. By assigning massive historical sequences to a linear attention branch and reserving a specialized softmax attention branch for recent interactions, our approach restores precise retrieval capabilities within industrial-scale contexts involving ten thousand interactions. To mitigate the lag in capturing rapid interest drifts within the linear layers, we furthermore design Temporal-Aware Delta Network (TADN) to dynamically upweight fresh behavioral signals while effectively suppressing historical ...
