Mapping Social Distress: A Computational Approach to Spatiotemporal Distribution of Anxiety

Abstract

Anxiety is a pervasive emotional state that tends to arise in situations involving uncertainty due partly to social and contextual issues including competition, economic disparity, and social insecurity. Thus, distribution of aggregate emotions, such as in anxiety, may reveal an important picture of otherwise invisible social processes in which individuals interact with local and global opportunities, constraints, and potential threats. The aim of this study is to present a computational approach to the dynamic distribution of anxiety extracted from natural language expressions of users of Twitter, a popular global social media platform. We develop an unsupervised machine learning procedure based on a naive Bayes model to classify contents of anxiety, estimate the degree of anxiety, and construct a geographic map of spatiotemporal distribution of anxiety. To validate our mapping results, a multilevel statistical analysis was performed to examine how anxiety distribution is correlated with other district-level sociodemographic statistics such as rates of birth and early divorce. Implications for further research and extension are discussed.