The development of the global tourism industry has driven increasing interest in cultural tourism recommendation algorithms. However, existing methods often rely on static modeling approaches for travel recommendations, neglecting users’ personalized and dynamic travel needs. This paper introduces a novel recommendation system, Spatio-Temporal Heterogeneous Information Network Collaborative Filtering (STHIN-CF). STHIN-CF integrates collaborative filtering with a two-layer graph network, where the static layer preserves long-term relational semantics without temporal noise, and the dynamic layer captures fine-grained user behavioral evolution without introducing structural bias. The subsequent fusion process, based on context-aware gating and bidirectional cross-attention, adaptively combines complementary information from both layers, effectively avoiding the information dilution commonly observed in unified spatiotemporal frameworks. By leveraging a spatiotemporally decoupled Graph Neural Network (GNN) framework, STHIN-CF jointly models the spatial and temporal characteristics of tourism resources. Experimental results demonstrate that STHIN-CF provides more accurate and adaptive cultural tourism recommendations than existing methods.
Han et al. (Tue,) studied this question.