This study proposes a hybrid recommendation algorithm integrating LDA topic model and collaborative filtering, aiming to improve the accuracy and diversity of cultural and creative product recommendations in tourist cities by combining semantic analysis and user behavior modeling. The LDA topic model is utilized to extract implicit topics from user comments and product descriptions, determine the optimal number of topics through confusion and consistency indicators, and quantify the distribution of user interest preferences and product features. And combined with collaborative filtering algorithm, the user-topic association matrix is constructed, and the dynamic recommendation effect is optimized by time weight (based on Ebbinghaus forgetting curve) and distance weight (minimum diameter circle method). The experimental part validates the model performance on three datasets, Ctrip, VW Dianping and Yelp, and the RMSE of this paper’s model on Ctrip dataset is 0.804, MAE is 0.752 and R-squared is 0.876 which are all better than the baseline models Caser, SLi-Rec and HGN, and on VW Dianping dataset, the RMSE’s 0.791 and MAE of 0.732 also perform best, verifying its robustness. In addition, the correlation analysis of user behavior shows that the correlation coefficient of 0.946 for payment behavior and 0.913 for order placing behavior are highly correlated with interest preferences. This study effectively mitigates the data sparsity and coldstart problem through the dual-path recommendation strategy and cluster filling technique.
