Emotional Landscapes: Mapping Urban Park Sentiments Using Natural Language Processing
Eric Delmelle, Xiayuanshan Gao · Abstracts of the ICA · 2025
Public parks are essential to urban life, fostering physical activity, mental health, and social cohesion.Traditional methods to assess park perception-like surveys-are often time-consuming and slow to adapt to urban change.This study introduces an automated, scalable approach for capturing public sentiment using geolocated Google Reviews and Natural Language Processing (NLP), applied to urban parks in Philadelphia.We show how emotional expressions vary across space and time, providing actionable insights for planners and community designers. Methodology.Our study analyzed over 37,000 Google Reviews from 2016 to 2024 across more than 200 parks in Philadelphia, each with a minimum of 25 reviews.Reviews were collected using Outscraper and processed with a finetuned RoBERTa model to classify 28 emotions, later grouped into six overarching clusters: Inspiration, Relaxation, Engagement, Discovery, Frustration & Annoyance, and Sorrow.We applied exponential decay weighting to favor recent reviews and geocoded the outputs for spatial analysis and visualization.Unlike many prior studies that focus only on sentiment polarity, we emphasize emotional richness and spatial storytelling through cartographic outputs.Emotions were visualized using choropleths, bivariate maps, and cluster-based classifications at the park level.Figure 1 outlines the analytic workflow.Figure 1.Workflow: Data collection, emotion classification, and spatial mapping.Results.The analysis revealed generally positive sentiments toward parks in Central, West, and Northwest Philadelphia, with temporal fluctuations influenced by seasonal factors.Positive sentiment peaked in 2021, likely reflecting increased