Multimodal Social Media Analysis Reveals Emotional Diversity in Urban Green Spaces

Urban green spaces (UGSs) are essential for human health and well-being. Data limitations in rapidly growing megacities such as China prevent assessing the mental benefits that nature can have on people, especially differential emotion types. Here, we develop a novel deep learning model capable of recognizing 4 distinct emotions from social media texts, i.e., happiness, relaxation, boredom, and anxiety, surpassing traditional binary sentiment classifiers. Leveraging co-occurring posts and images from 148,720 UGS users on the Sina Weibo platform in Xiamen, China, we investigated the dynamic correspondence between UGS features and the associated emotions. We find that happiness is positively linked to sky and blue views but negatively to building views, while relaxation is associated with green and colorful views. Both boredom and anxiety decline with increasing park area, especially in parks smaller than 200 ha. Small, nearby, and green parks are conducive to relaxation, while large parks with expansive skies and blue views are more effective for recharging. This artificial-intelligence-driven framework provides a useful tool for tracking nature-related emotions, which can be extended to other fast-growing urban areas. These findings offer a general reference for human-centered urban planning and development elsewhere in the world.