Characterization of Behavioral Transitions Through Social Media Analysis: A Mixed-Methods Approach

Unhealthy behaviors are a socioeconomic burden and lead to the development of chronic diseases. Relapse is a common issue that most individuals deal with as they adopt and sustain a positive healthy lifestyle. Proper identification of behavioral transitions can help design agile, adaptive, and just-in-time interventions. In this paper, we present a methodology that integrates qualitative coding, machine learning, and formal data analysis using stage transition probabilities and linguistics-based text analysis to track shifts in stages of behavior change as embedded in journal entries recorded by users in an online community for tobacco cessation. Results indicate that our semi-automated stage identification method has an accuracy of 90%. Further analysis revealed stage specific language features and transition probabilities. Implications for targeted social interventions are discussed.