Travel Recommendation Model Integrating Long-term and Short-term User Preferences

Personalized travel recommendations are very important for users to find attractions of interest and improve users’ travel experience. A key issue in travel recommendation systems is to accurately learn vector representations of users to capture their preferences. Users often have both long-term and short-term preferences. However, existing travel recommendation algorithms usually only learn a single representation of the user, which may not be enough. In this paper, we propose a travel recommendation model that integrates users’ long-term preferences and short-term preferences (TRLS), which can learn the user’s long-term preferences and short-term preferences from the user’s historical behavior sequence. Specifically, the model includes two core modules, namely tourist attraction encoder and user encoder. Among them, the attraction encoder learns a unified representation of the attraction from the attraction introduction and comment title. In the user encoder, the BST model and the BI-GRU model are used to learn the user’s long-term preference representation and short-term preference representation respectively. In order to improve the quality of learning, we further use the attention mechanism to fuse the two preferences. We have conducted a large number of rating prediction experiments on real data sets. The results show that compared with previous models, our proposed model has advantages in evaluation indicators such as RMSE and MAE, which proves the effectiveness of our proposed model.