Identifying Social Experience Typologies across Urban Parks from User-Generated Reviews

Measuring how different urban parks support social interaction across a city remains constrained by data limitations and the high cost of traditional surveys and observations. Urban parks are essential settings for social interaction and community life, yet their social benefits vary widely within cities and are unevenly distributed across neighbourhoods. This study introduces a transformer-based classification approach to measure social experience across diverse urban parks, analyzing Google Reviews from 154 urban parks in Vancouver, Canada. We fine-tuned RoBERTa transformer models to classify social interaction content within 49,563 reviews, deriving three park-level indicators: volume of reported social experience, affective tone, and overall satisfaction. This supervised approach advances beyond general sentiment analysis by isolating social dimensions within user-generated content. Latent profile analysis reveals four park social experience typologies: quiet neighbourhood parks (34%), struggling spaces (3%), community hubs (39%), and high-performing parks (24%). Multinomial logistic regression demonstrates that park size and amenity diversity are more consistent predictors of typology membership than neighbourhood demographics. Struggling spaces, concentrated in Vancouver’s Downtown Eastside, combine high social-review volume with negative reported experiences, showing that volume and affective tone do not necessarily coincide. These findings demonstrate how transformer-based analysis of user-generated content enables scalable, citywide assessment of park social experience, offering municipalities a more nuanced and targeted basis than volume-based indicators alone for directing design and management investment.