<p>Cross-domain sequential recommendation (CDSR) aims to predict user’s next interactions by analyzing their sequence of past behaviors across different domains. Existing methods rely heavily on overlapping users, limiting their practical applicability. To address this issue, this paper proposes an Attention and Temporal-Aware Cross-Domain Sequential Recommendation (ATA-CDSR) model, which effectively captures both single-domain and cross-domain user preferences. Our approach is built upon a graph-based framework consisting of three core modules: (1) Single-domain sequence graphs are constructed for all users to capture temporal-aware sequential patterns; (2) A cross-domain sequential graph with a dual attention mechanism (node-level and domain-level) is built to transfer collaborative signals across domains; and (3) A temporal-aware encoder is introduced to model relative time intervals for capturing user’s dynamic preferences. Experimental results on two real-world datasets show that ATA-CDSR significantly outperforms state-of-the-art methods, achieving an aggregated average improvement of 13.8% over the strongest baseline on NDCG@10 and HR@10.</p>