Multi-cluster data mining and analysis of tourist behaviour patterns for scenic area management

To address the fragmentation in tourist need identification and the disconnect between a multi-model fusion analysis method is proposed.This approach uses a bidirectional long short-term memory (Bi-LSTM) network to extract semantics from review texts and a latent Dirichlet allocation (LDA) model to identify core topics.A spatiotemporal cube structure maps emotional labels to spatiotemporal coordinates, quantifying experiential differences and optimising tourist group segmentation.Experimental results showed that five themes from both positive and negative comments and five from negative comments were well-separated, effectively reflecting dimensional differences in tourist feedback.Multiple regression models indicated varied group preferences, with one group favouring architectural features (preference coefficient of 1.820) and another prioritising affordability (preference coefficient of 2.186).The overall prediction accuracy of the model is 0.82.The research results provide data-driven decision-making basis for precise service design and resource optimisation allocation in scenic spots.