Exploring geo-tagging behaviour in social media data through structural topic modelling and geographically weighted regression

In this study, Tweets were used as inputs for structural topic modelling (STM) to detect what topics were expressed by social media users. The STM identified 20 topics ranging from daily life to professional services. The results revealed how topic words were associated with geo-information versus non-geo-information in their posts. Additionally, geographically weighted regression (GWR) was used to investigate how geo-topics were spatially associated with land-use types, the results identified the specific areas where the geo-topics were positively correlated with different land-use categories.To avoid privacy concerns and to follow the terms of Twitter/X API, the raw individual Tweets are not published. The API terms here (https://developer.x.com/en/more/developer-terms/agreement-and-policy).

本研究以推文（Tweets）作为结构主题模型（Structural Topic Modelling，STM）的输入数据，用于识别社交媒体用户所表达的主题。STM共识别出20类主题，覆盖从日常生活到专业服务的广泛范畴。研究结果揭示了推文中的主题词汇与地理信息（geo-information）、非地理信息之间的关联模式。此外，本研究采用地理加权回归（Geographically Weighted Regression，简称GWR），探究地理主题与土地利用类型的空间关联特征，最终识别出地理主题与不同土地利用类别呈正相关的具体区域。为规避隐私风险并符合推特/X应用程序编程接口（Twitter/X API）的使用条款，本研究未发布原始单条推文数据。相关API使用条款可参阅：https://developer.x.com/en/more/developer-terms/agreement-and-policy。