.png){kind=icon}
.png){kind=icon}
Return
Causality-aware dual-scale preference model for sequential recommendation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.131484.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Sequential recommendation has attracted considerable attention in recommendation systems for its natural ability to capture the dynamics of user behaviors. Although multi-scale causal methods have achieved notable progress in modeling, most existing approaches fail to effectively exploit valuable bias information. As a result, the debiased representations often lose critical personalized signals, which in turn limits recommendation performance.To address this issue, we propose a Causality-Aware Dual-Scale preference model for Sequential Recommendation (CADSR), which collaboratively models users’ long-term stable preferences and short-term personalized interests. Specifically, CADSR constructs a novel causal graph for sequential recommendation to explicitly characterize causal pathways and the influence of confounding variables. It then applies a front-door adjustment strategy to remove bias from long-term preferences, while retaining and amplifying the key bias signals embedded in short-term interactions to enhance personalization.Extensive experiments on five real-world datasets demonstrate that CADSR achieves average improvements of 4.55% in HR@5 and 5.13% in NDCG@5 over state-of-the-art sequential recommendation methods, confirming its overall superiority in terms of recommendation accuracy, robustness, and personalized modeling capability.
Keywords:
Sequential recommendation
Causal modeling
Dual-scale preference
Bias removal
Personalized modeling
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W