Understanding Intrusive Questioning in Online Conversations

Abstract

In online mental health platforms, ensuring that vulnerable users receive care without harm is of paramount importance. Our group created a representative, data-driven semantic taxonomy for harmful behaviors that occur on these platforms and then further defined and analyzed one of these behaviors: intrusive questioning, which we found was linked to negative experiences on the platform. Then, after labeling 100k+ messages, we created a model that could identify intrusive questioning with a high accuracy that surpassed a baseline keyword approach.