Machine Learning Insights into Eating Disorder Twitter Communities

Abstract

Eating disorders have serious impacts on young population physical, psychological, and social functioning. Health agencies are calling for new psycho-therapeutic intervention considerations that take into account online communities, which is not possible if therapists know little about eating disorders discussions online. In this paper, we leverage machine learning analytics to understand what the eating disorder communities are talking about over time and how they are talking about it. By analyzing local and global community discussions, we discovered complex group dynamics underpinning collective identities offering emotional support but also potentially perpetuating harmful behaviors. Our analysis of four local subcommunities and four global theme evolutions revealed prevalent subjects, perspectives, motivations, and linguistic patterns. We found tight-knit communities grounded in shared membership, goals, cultures, and practices. Community voices highlighting recovery journeys were limited. Our computational assessment of invisible online spaces aims to inform personalized interventions accounting for community forces in youth mental health.