Stassinos James, Anderson Taylor, Züfle Andreas
Department of Geography and Geoinformation Science, George Mason University, Fairfax, Virginia, United States of America.
Department of Computer Science, Emory University, Atlanta, Georgia, United States of America.
PLoS One. 2025 Jun 2;20(6):e0325166. doi: 10.1371/journal.pone.0325166. eCollection 2025.
Social media data has shown potential for identifying infectious disease outbreaks faster than official records of disease incidence. We examine spatial, temporal, and spatiotemporal relationships between COVID-19-related microblog sentiment and COVID-19 cases over space and time to investigate whether microblog-derived sentiment can be used for local infectious disease outbreak early warning. Therefore, we measure the sentiment of 56,755,894 COVID-19 related microblogs (tweets) from the microblogging platform X. We group these tweets by county and by calendar week to investigate spatial and temporal correlation between sentiment and observed cases (in the corresponding county and week). Our temporal analysis shows a significant negative correlation between sentiment and cases between June and September 2020. During this time, tweet sentiment could have served as an early warning for new COVID-19 outbreaks. Our spatial analysis shows that the East of the United States exhibits a significant negative correlation between Sentiment and Cases while the West exhibits a significant positive correlation. In these regions, Tweet Sentiment could have been used as an early warning signal for new outbreaks. Our spatiotemporal analysis discovers even stronger correlations in certain regions during certain time periods. If we could understand when, where, and why this correlation is strong, then we may be able to leverage social media as a successful early warning system.
社交媒体数据已显示出比官方疾病发病率记录更快识别传染病爆发的潜力。我们研究了与COVID-19相关的微博情绪与COVID-19病例在空间和时间上的空间、时间和时空关系，以调查源自微博的情绪是否可用于本地传染病爆发的早期预警。因此，我们测量了来自微博平台X的56755894条与COVID-19相关的微博（推文）的情绪。我们按县和日历周对这些推文进行分组，以研究情绪与观察到的病例（在相应的县和周）之间的空间和时间相关性。我们的时间分析表明，2020年6月至9月期间，情绪与病例之间存在显著的负相关。在此期间，推文情绪本可作为新的COVID-19爆发的早期预警。我们的空间分析表明，美国东部地区情绪与病例之间存在显著的负相关，而西部地区则存在显著的正相关。在这些地区，推文情绪本可作为新爆发的早期预警信号。我们的时空分析发现，在特定时间段内某些地区的相关性更强。如果我们能够了解这种相关性何时、何地以及为何强烈，那么我们或许能够将社交媒体作为一个成功的早期预警系统加以利用。
