Machine-Reading Places & Spaces: Generative Probabilistic Modelling Of Urban Thematic Zones & Contexts
In this paper, a “place” is conceptualised as a composition of dynamic socioeconomic activities and collective perceptions. We apply generative probabilistic modelling to explore urban contextual semantics. By analogy to sorting documents into different topics, this research retrieves data embedding for each urban regions and classify them with thematic zones. Using Singapore as a case study, topic modelling is applied to retrieve perceptual and functional thematic zones from Instagram and TripAdvisor respectively. A subsequent analysis shows strong correlations among certain regions with functional and perceptual consistency. In addition, with our proposed uniqueness and diversity indices, a strong negative correlation at 0.82 is found, suggesting that a region could be more unique if the functions tend to be dominated by certain types of functional and perceptual thematic zones.
Keywords: Machine Learning, Natural Language Processing, Generative Probabilistic Models, Urban Data Modelling, Thematic Zones, Topic Modelling, Sdg 11