Hybrid LSTM-Transformer Model for Sequential and Context- Aware Tourism Destination Recommendation

Abstract

With the vigorous development of big data and artificial intelligence technology, personalized recommendation systems have been deeply integrated into the tourism industry, aiming to improve user experience through accurate travel recommendations. However, traditional recommendation algorithms often struggle to capture the dynamic changes of user interests and long-tail preferences when dealing with complex sequential data. To this end, this study constructs a hybrid architecture that integrates long short-term memory (LSTM) and transformer to build an efficient tourist destination recommendation system. Among them, the LSTM module focuses on mining the time-dependent features of user behavior, and the Transformer captures the long-range dependencies in behavior by relying on the self-attention mechanism. The experiment was conducted based on a real dataset of more than 100 million user browsing records, and the results showed that the prediction accuracy of the system was improved by 23% and the recall rate was increased by 18% compared with the baseline method. User feedback analysis further showed that the new system's recommendations were more relevant to actual needs, and the average satisfaction score increased by 15%. These results not only confirm the great potential of deep learning in optimizing the personalized recommendation system for tourism, but also provide key technical support for the intelligent upgrading of the tourism industry.