Scale-Dependent environmental influences on urban green space sentiment: Integrating multimodal social media analysis and explainable spatial models

The benefits of urban green spaces (UGSs) to public health arise not only from the green spaces themselves but also from the surrounding environment (SE). However, leveraging multimodal social media data to comprehensively assess the associations between environmental features and sentiment toward UGSs remains challenging. This study integrates SHapley Additive exPlanations and Geographically Weighted Regression, using 59,880 social media text entries and 49,501 images from 280 UGSs in Beijing to reveal nonlinear associations, synergies, and spatial heterogeneities across different scales of UGSs. The results show that: (1) SE features play a more important role than UGS attributes in determining UGS sentiment, with building coverage ratio, gross domestic product, and population density contributing the most. (2) All environmental variables exhibit nonlinear, interactive effects, and geographic heterogeneity on UGS sentiment. (3) Within SE features, accessibility is positively associated with sentiment in community green spaces but negative in non-community green spaces, whereas floor area ratio is positive in both, revealing scale-dependent heterogeneity across UGSs and localized effects within their effective ranges. Our methodology combines photo- and text-based sentiment analysis, offering a more accurate and efficient approach to capturing public insights and thereby enabling more precise UGS planning decisions.