Identifying drug and alcohol displays on social media using a machine learning approach, and mechanisms that impact adolescent substance use

Adolescent substance use can result in several challenges and long-term consequences, including academic difficulties, injuries, and a substance use disorder. Social media provides adolescents with opportunities to share and view a variety of substance use behaviors in the absence of negative consequences and may now represent a key socialization mechanism that has not been widely investigated. This dissertation is comprised of three studies that focus on the impact of engagement in social media on adolescent substance use using two different methodologies. First, ethical considerations specific to adolescents and young adults in social media research will be outlined. Then, secondary data analyses that assessed whether subjective norms mediated associations between exposure to alcohol and marijuana-related content posted by peers and influential figures on Instagram and Snapchat and offline adolescent alcohol and marijuana use will be presented. Findings indicate that increased exposure to online substance use content was more consistently associated with injunctive norms rather than descriptive norms. Lastly, a proof of concept study highlights how machine learning can be used to detect adolescents engaging in e-cigarette use by their Instagram metrics. Overall, the supervised learning model had good performance and can be used to detect adolescent e-cigarette use. The collection of work presented in the dissertation emphasizes the potential impact that social media can have on adolescent risk behaviors, like substance use. Future research should examine how specific types of engagement, as well as features, on social media can impact adolescent substance use and well-being.

Major Professor: Dr. Elisa M. Trucco