Investigating the dynamicity of sentiment predictors in urban green spaces: A machine learning-based approach

Abstract

Urban green spaces (UGSs) are known to promote psychological restoration. However, most traditional studies measure this effect statically using preset assumptions or scenarios, and tend to neglect factors’ varied effect under different conditions. In this study, we investigated the dynamicity of UGS-related factors on human sentiment under different UGS sizes and seasons by analysing social media reviews using machine learning. To this end, first, self-reported online reviews of UGSs in Nanjing City were collected. Next, machine learning-based text mining and natural language processing were performed to extract the latent topics and sentiment level of each review document, respectively. A logistic regression model series was established to analyse the influential valence of the review topics on the sentiment of the reviews. The results revealed that people’s sentiments and corresponding predictor performances are more sensitive to variations in UGS sizes compared with the seasons. Within UGSs, natural and activity features play more important roles in promoting positive sentiments compared with cultural and facility features. Moreover, larger UGSs include more tools that promote positive sentiments, and waterside activities and daily exercise promote positive sentiments throughout the year. Compared with previous studies, this study adopted a dynamic perspective to examine the influence of UGSs on people’s sentiment and provides a new approach in environmental psychoanalysis by integrating different ML-based techniques. Our findings are expected to aid in future decision-making during UGS planning and design in Nanjing and similar cities worldwide.