URBAN SENTIMENT MAPPING USING LANGUAGE AND VISION MODELS IN SPATIAL ANALYSIS

Jayedi Aman ^* Timothy C. Matisziw

• University of Missouri, Columbia, United States

This study explores the relationship between urban environments and public sentiment by evaluating social media data with respect to features extracted from street view imagery.Leveraging language and vision models, the research analyzes the extent to which different urban features may influence public emotions across various spatial contexts and timeframes. To this end, a BERT-based transformer model is used to extract characterizations of public sentiment from geotagged social media posts. Then, computer vision models, such as PSPNet and Mask R-CNN, are used to quantify qualities of urban design like visual enclosure, human scale, and streetscape complexity. By integrating publicly available data and spatial simulation techniques, the study reveals how the between urban features and sentiment can change over time. The findings indicate a positive influence of features such as greenery and pedestrian-friendly infrastructure on sentiment and provide insight as to how societal disruptions may impact sentiment The research underscores the importance of incorporating public sentiment into territorial and spatial-based simulations to foster more inclusive, safe, and resilient environments.