What Topics Matter? Applying Structural Topic Modeling (STM) to Extract Insights From Social Media-Based Health Interventions

• In Event: Health Communication and Social Media (High-Density)

Abstract

Recent meta-analyses of social media-based health interventions have shown considerable heterogeneity in effectiveness. Textual data on intervention-induced discussion has been underutilized to understand why some interventions would succeed and some fail, partially due to the lack of a scalable statistical model to link textual data with behavioral outcomes. By analyzing discussion data from the Birth Control Connect (BCC) online intervention that aims to use social influence to promote intrauterine devices (IUD) among young women, we illustrate the usefulness of a new statistical text analytic tool, Structural Topic Modeling, in estimating treatment effects on topic prevalence and automatically identifying important topics that mediate treatment effects on known psychological precursors to actual adoption of IUD. We demonstrated with evidence that STM can help researchers to not only automatically extract discussion topics that are sensitive to the intervention, but also to identify critical topics that link treatment to behavioral outcomes in nuanced ways.

Authors

• Sijia Yang, U of Pennsylvania

• Jingwen Zhang, U of California, Davis

• Christine Dehlendorf, U of California - San Francisco

• Damon Centola, U of Pennsylvania