Using Topic Modeling to Detect and Describe Self-Injurious and Related Content on a Large-Scale Digital Platform

Objective: Self-injurious thoughts and behaviors (SITBs) are a complex and enduring public health concern. Increasingly, teenagers use digital platforms to communicate about a range of mental health topics. These discussions may provide valuable information that can lead to insights about complex issues like SITBs. However, the field of clinical psychology currently lacks an easy-to-implement toolkit that can quickly gather information about SITBs from online sources. In the present study, we applied topic modeling, a natural language processing technique, to identify SITBs and related themes online, and we validated this approach using human coders. Method: We separately used topic modeling software and human coders to identify themes present in text from a popular online Internet support forum for teenagers. We then determined the degree to which results from the software's topic model aligned with themes identified by human coders. Results: We found that topic modeling detected SITBs and related themes in online discussions in a way that accurately distinguishes between relevant and irrelevant human-coded themes. Conclusions: This approach has the potential to drastically increase our understanding of SITBs and related issues discussed on digital platforms, as well as our ability to identify those at risk for such outcomes.