This paper proposes a smart tourism personalized recommendation algorithm based on Deep Attention Interest Collaborative Filtering (DAICF), which solves the sparsity and cold-start problems in tourism recommendation by fusing geo-tagged photo data with deep learning techniques to mine explicit and implicit relationships of user interests. The research utilizes P-DBSCAN clustering algorithm to identify hotspot locations, constructs user-tour route binary association matrix, and designs deep collaborative filtering model, combining shallow linear interaction with deep nonlinear interaction, to synchronously mine the low-order and high-order relationships between user and item features. Based on the real dataset Yelp, DAICF performed well in the Top-N recommendation task: the Precision@5 was 20.13% and the Recall@15 was 35.26%, which was significantly higher than that of the baseline models such as EPT-GCN and ASGNN. In addition, the ablation experiments show that key components such as social relations and temporal context contribute prominently to the model performance. In practical application tests, the DAICF-generated travel routes shorten the total distance of the trip by 52.5% and improve the time-to-time ratio by 41.9%, which verifies its efficiency and practicality in trip planning.