Emotional landscapes: Mapping urban park sentiments using natural language processing

Source abstract

Urban parks play a critical role in promoting physical activity, mental health, and community well-being. However, understanding how these spaces are perceived by the public remains difficult, as traditional surveys are costly, slow to update, and limited in their geographic coverage. This study presents a novel analytical framework that integrates transformer-based emotion classification, dimensionality reduction, clustering, and spatial autocorrelation to monitor park sentiments and emotions at scale, analyzing over 37,000 Google Reviews of 93 parks in Philadelphia. We extract 28 nuanced emotions, later reduced to six distinct categories (e.g., inspiration, frustration) via principal component analysis and clustering. These emotion profiles are mapped spatially and temporally, revealing strong geographic variation and seasonal trends. A time-weighted approach prioritizes recent reviews, allowing planners to identify shifting perceptions tied to maintenance, safety, or investment. Finally, we link emotional patterns to park amenities and surrounding socioeconomic conditions, offering actionable insights for park planning. Our findings suggest that natural language processing and emotional cartography can improve adaptive park management, spatial equity, and real-time public participation in landscape planning.