Understanding park-based health-promoting behavior and emotion with large-scale social media data: The case of Tianjin, China

Urban parks' positive impacts on human physical, mental, and social well-being have been widely discussed in urban planning and public health literature. As a result, scholars have invested considerable efforts in measuring park-based health-promoting behavior and emotion (e.g., physical activity, social interaction, and positive emotion). Recent years have seen a rapidly growing trend of using publicly available social media data as a potential measurement of park-based behavior. However, most ongoing studies focused on visitation amounts and paid little attention to finer-grained visitor behavior. In this study, we proposed a machine learning-aided text mining method to extract detailed park-based behavior using large-scale social media data, in this case, reviews posted to the Dianping.com social media platform. Our approach combined manual labeling with machine learning-based natural language processing models to leverage the accuracy of manual coding and the efficiency introduced by computer-aided tools. As a proof of concept, we applied the proposed method to 23,910 park-related online reviews in Tianjin, China, revealing the widely heterogeneous health-promoting behavior and emotion reported in 34 urban parks in downtown Tianjin. To our knowledge, this is the first available data set of park visitors' behavior for the city. This study shows how scholars and practitioners can turn unstructured social media data into structured urban behavior insights that are helpful for making better places. Moreover, we discussed the limitations of the current approach and future research efforts to validate and improve it.