Detecting Eating Disorders From Social Media Content: What Has Been Done and Where Do We Go Next?

Abstract

Objective: Less than 20% of individuals with eating disorders (EDs) ever receive treatment, highlighting a need for scalable, innovative methods of identifying and providing support to individuals with ED symptoms. At the same time, ED-related content on social media (SM) platforms is pervasive, offering an opportunity to detect signals of ED symptoms from SM data. This paper examines how artificial intelligence (AI) and computational methods can be leveraged to detect ED symptoms from SM content and provide timely intervention.

Method: We review SM-based ED detection methods researched to date, including content tags, topic modeling, and natural language processing. We also discuss critical next directions for this area, including opportunities to pair detection with digital interventions, and examine challenges in developing, evaluating, and implementing these tools. Finally, we offer recommendations for ED experts for guiding the development, evaluation, and deployment of robust detection systems.

Results: Research supports the feasibility of harnessing SM data to identify individuals with ED symptoms and has begun exploring methods of pairing SM-based ED detection with interventions. Although SM platforms already use automated methods of detecting and moderating harmful content, these systems are not transparent and show room for improvement, highlighting the importance of ED experts' involvement in developing detection methods.

Discussion: Leveraging SM data presents an unprecedented opportunity to identify and provide support to millions of individuals with ED symptoms. Research, interdisciplinary collaborations, and ethical safeguards can transform SM into a supportive resource for individuals with EDs.