The last decade has seen explosive growth of online social networks (OSNs), which are now a pervasive part of everyday life for many people. Big data analyses of OSNs have enormous potential to enable new discoveries and applications in public health. For example we may soon be able to study how mental health indicators vary over time and between communities and regions; how major events like disasters and mass shootings affect community mental health; and potentially even use social media for individual-level screening and diagnosis. But in order to do this effectively, psychologists need to know how to make sense of vast amounts of digital records of raw behavior. That will require interdisciplinary collaborations between psychology and computer science to create new tools and methods for automated analyses of mental health variables in large and complex datasets. The goal of this project is to develop new metrics for doing large-scale, big data mental health research on Twitter, a popular online social network service with a large and diverse user base in the United States that makes most user-generated content publicly available. We are particularly interested in how Twitter data can be used to draw inferences about users' personality traits (which are indicative of risk and vulnerability), emotions, and clinical symptoms. To date, researchers studying mental health on Twitter have focused on a small number of metrics which often differ from study to study and are not always well-validated. We do not yet have a unified, comprehensive understanding of how mental health characteristics can be measured and studied online. Our project has three goals. First, we will gather data from a very large, representative sample of American Twitter users to map out what meaningful metrics can be derived from Twitter data. What can be reliably measured, and how are different kinds of metrics related to each other? By studying relationships and overlaps among these variables, we expect to identify a key set of online metrics that can be used to characterize important psychological differences among users. Second, we will recruit a sample of Twitter users to complete standardized assessments of personality, emotion, and clinical symptoms. We will look to see what kinds of online behaviors are reliable and valid markers of these psychological variables. Third, to demonstrate the utility of these markers we will monitor millions of Twitter accounts around the country for one year to measure how one specific kind of community trauma, mass shootings, affects indicators of personality, emotion, and mental health. We expect to find effects of such events on the online expression of emotion and clinical symptoms immediately following an event, and perhaps longer-term effects that endure beyond the immediate aftermath.
This project will allow us to identify important markers of personality, emotion, and mental health in Twitter data. This will be useful to researchers who want to use 'big data' approaches to study mental health, and to policy applications to study individual or regional variations and reactions to events.
