Work

Co-LS: decoupling collaborative awareness and long–short-term interests modeling for personalized tourism recommendation

Year: 2026
Type: article
Abstract: Understanding and predicting tourist mobility is central to location-based services, as it reveals how tourists interact with places under spatial, temporal, semantic, and behavioral constraints. Existing recommendation models often suffer from semantic misalignment between macro-level collaborative mobility structures and micro-level individual sequence dynamics, making them vulnerable to data sparsity. To address this issue, we propose Co-LS, a unified framework that decouples group-level collaborative structure learning from individual long- and short-term interest modeling for personalized attraction recommendation. Specifically, Co-LS constructs a multi-relational heterogeneous collaborative graph enriched with tourism prior knowledge to model multidimensional group mobility constraints. A collaboration-aware graph neural network captures high-order topological dependencies while reducing noisy information propagation. Meanwhile, a hierarchical recurrent architecture with local–global attention models both stable long-term preferences and dynamic short-term travel intentions. To bridge heterogeneous representation spaces, cross-view contrastive learning encourages self-supervised consistency between collaborative and individual preference views. Finally, an attention fusion module adaptively integrates aligned multi-source representations for next-attraction prediction. Experiments on real-world tourism datasets from Beijing and Shanghai show that Co-LS consistently outperforms representative baselines. These results demonstrate its effectiveness for tourist mobility prediction and its potential applicability to personalized tourism services.
Language: English
Open Access status: gold
Estimated APC paid: $1,990
OpenAlex ID: W7236858800
FWCI:
Cites: 51
Cited by:
Related to: 10