TME-PSR: Time-aware, Multi-interest, and Explanation Personalization for Sequential Recommendation

Sequential recommendation systems aim to predict user preferences based on historical interactions, but most methods still fall short in delivering comprehensive personalization. We propose TME-PSR, a unified framework that simultaneously addresses three dimensions of personalization: (1) Time-aware personalization — a dual-view gated time encoder that adaptively captures personalized short-term and long-term temporal rhythms, pluggable into existing models without architectural changes; (2) Multi-interest personalization — a lightweight multihead Linear Recurrent Unit (LRU) architecture that automatically disentangles multiple fine-grained sub-interests with improved efficiency; and (3) Explanation personalization — a dynamic dual-branch mutual information weighting mechanism that adaptively adjusts the semantic alignment between recommendations and explanations for different users. Extensive experiments on three real-world datasets demonstrate that TME-PSR consistently improves both recommendation accuracy and explanation quality over baselines, with better computational efficiency.

Recommended citation: Qingzhuo Wang, Leilei Wen, Juntao Chen, Kunyu Peng, Ruiyang Qin, Zhihua Wei, Wen Shen. (2026). "TME-PSR: Time-aware, Multi-interest, and Explanation Personalization for Sequential Recommendation." arXiv 2026.
Download Paper