From Narratives to Destinations: Semantic–Spatial Modeling of Tourism Trends Using Geotagged Reviews

Abstract

Tourism narratives on social media and review platforms contain rich semantic and spatial information that can reveal evolving tourist preferences and destination trends. This study presents a hybrid topic modeling framework that integrates semantic embedding-based topic extraction with spatial-temporal clustering to analyze geotagged TripAdvisor reviews from 2019 to 2024. We employed state-of-the-art natural language processing techniques, including BERT-based topic modeling and Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction, to uncover nuanced themes and their geographic distributions. Spatial entropy, temporal prevalence, and Moran’s I statistics were used to characterize spatial coherence and temporal evolution of dominant themes. Results indicate that our hybrid model significantly outperforms traditional Latent Dirichlet Allocation (LDA) in topic coherence (0.581 vs. 0.513), while spatial clustering reveals meaningful patterns in eco-tourism, cultural heritage, and health-safety topics. Temporal shifts highlight a post-COVID transition from budget-consciousness to experience- and sustainability-driven narratives. The proposed framework provides destination management organizations (DMOs) with a powerful tool for geographically and contextually targeted tourism analytics. This study contributes methodologically by aligning semantic spaces across locations and temporally modeling evolving themes, paving the way for more intelligent, data-driven tourism planning.