Improving Spatio-Temporal Topic Modeling with Swarm Intelligence

Abstract

This study introduces innovative methodologies for spatiotemporal topic modeling applied to the TripAdvisor forum of Morocco, leveraging the diverse and geographically tagged user-generated content. We develop and evaluate two schemas integrating Latent Dirichlet Allocation (LDA) with advanced clustering techniques, including a hybrid K-Means algorithm that incorporates Genetic Algorithms and the Artificial Bee Colony method. The first schema independently processes user threads, publication times, and locations using LDA, followed by clustering, while the second schema combines these dimensions into a unified vector for holistic LDA application, facilitating direct comparisons of clustering efficacy. Our findings demonstrate that swarm intelligence significantly boosts clustering performance, especially for larger clusters, and enhances the visualization of complex data relationships. These insights offer actionable intelligence for tourism stakeholders and underscore the practical benefits of advanced computational techniques in harnessing user-generated content for strategic decision-making.