About

This project will identify interaction patterns with online video platforms that are indicative of suicide risk, focusing on YouTube and TikTok. Leveraging archival data including over 5 million interaction events collected from participants in previous research, we will use combinations of neural language models to identify suicide-related “like”, “search” and “watch” events. We will then assess the temporal relationships between suicide-related interaction events and suicidal ideation, behavior and mental health challenges reported by these participants. Building on these analyses, we will proceed to model patterns of interaction, differentiating between user-initiated (e.g. search) and algorithm-prompted (e.g. recommended content without a preceding search) content to characterize the ways in which intentional and algorithmically-driven behavior drive exposure to suicide-related content. In addition, we will develop a prototype of a privacy-preserving risk monitoring tool, which will detect interactions with concerning content and leverage light-touch intervention strategies to mitigate its impact.

Outcomes

In progress.