Tourism consumer behavior prediction and personalized product recommendation based on deep learning and big data

The growth of online travel platforms necessitates sophisticated recommendation systems that can simulate fluctuating tourist behavior and produce actionable travel itineraries. Existing tourism recommender systems face significant challenges, including inadequate modeling of sequential and spatio-temporal dependencies, reliance on static user preferences, and the inability to generate constraint-aware itinerary-level recommendations within time, budget, and distance constraints. This paper introduces a unified deep learning network that leverages big data techniques to build the TourRec-Stack framework, which combines spatio-temporal sequential representation learning, multi-task behavior prediction, personalized ranking, and constraint-aware itinerary optimization to predict tourism consumer behavior accurately. Using the Foursquare NYC and Tokyo check-in dataset, user interaction sequences are encoded into dynamic embeddings and processed through a spatio-temporal network to predict next-POI visits and visit likelihood. Then, the predicted intent is used to make personalized rankings and possible itineraries. Experimental results show that TourRec-Stack is 8–12% better than the best baselines in Precision@10, 7–10% better in Recall@10, and 9–14% better in NDCG@10. It also raises AUC by about 6–9%. The framework also reduces inference latency by almost 15% compared to the strongest baseline. These results show that TourRec-Stack is a good choice for intelligent tourism recommendation systems and it is accurate and scalable.