Using natural language processing to understand changes in emotions and psychopathology: Evidence from observational cohorts and a clinical trial

Introduction: Natural language processing (NLP) metrics have emerged as a way of understanding emotions. We explored the application of NLP to understand emotion and emotion regulation in psychopathology in two studies In Study 1, we analyze social media messages from the Studies of Online Cohorts for Internalizing symptoms in Language (SOCIAL), two observational cohorts of social media users (N=3,334). In Study 2, we analyze data from a cognitive restructuring task embedded in the cognitive-behavioral therapy condition (n=409) of a large treatment trial (N = 828). Methods: In Study 1, we obtained Valence Aware Dictionary for sEntiment Reasoning (VADER) scores for 1,471,079 million Tweets. We used median VADER scores and variability in VADER scores to predict different dimensions of psychopathology. In Study 2, we use VADER and Sentiment Analysis and social Cognition Engine (SEANCE) to analyze 28,986 words from the cognitive restructuring task to assess whether NLP metrics are indicators of emotion regulation in the task, are correlated to individual differences, and can predict positive (well-being) and negative (depression) mental health at weeks 2 and 8. Results: In Study 1,affect variability, especially negative affect was associated with more severe Internalizing-fear and distress and Somatoform-distress symptoms and less severe Externalizing-substance use symptoms. In Study 2, NLP metrics improved during the cognitive restructuring task, were associated with self-report use of cognitive reappraisal, emotional stability, and extraversion, and predicted symptom change post-intervention. Discussion: These studies illustrate the potential to use NLP for understanding changes in emotional dynamics, including minute-scale changes in emotions during psychological interventions.