Natural language processing meets spatial time series analysis and geovisualization: identifying and visualizing spatio-topical sentiment trends on Twitter

ABSTRACTPrevious studies have introduced various approaches for visualizing the spatial and temporal distributions of sentiments expressed on social media. However, many existing methods either overlook the evolving nature of sentiments or fail to account for the spatial distribution of sentiment trends related to specific topics. To gain a comprehensive understanding of how sentiments evolve in relation to topics and geographies, it is essential to capture the dynamic nature of sentiment through time series analysis and geovisualization. This article introduces a workflow that combines natural language processing, spatial time series analysis, and geovisualization techniques to identify and visualize the variations in sentiment trends on Twitter across different geographic regions and topics. By examining the 2016 presidential debates as a case study, we uncover distinct temporal patterns in sentiment distributions across various topics and states. Our findings also show that adjacent states do not always share similar sentiment trends, and that geographic clusters with similar sentiment trends also vary across topics. Failing to consider these variations may result in misunderstanding public discourse and sentiments since they are diverse and dynamic in nature.KEYWORDS: Sentiment trendsnatural language processingspatial time series analysisgeovisualizationTwitter AcknowledgmentsThe authors would like to thank the anonymous reviewers for their valuable insights. The reviewers’ constructive comments greatly contributed to the improvement of this manuscript.Disclosure statementNo potential conflict of interest was reported by the author(s).Data availability statementTwitter data used in this study are openly available in figshare at https://doi.org/10.6084/m9.figshare.20277840.v1. The shared data contain tweet IDs related to a series of three presidential debates in 2016 between the dates of September 26 and 26 October 2016.Supplementary dataSupplemental data for this article can be accessed online at https://doi.org/10.1080/15230406.2023.2264751.