EDTok: A Dataset for Eating Disorder Content on TikTok
Charles Bickham, Bryan Ramirez-Gonzalez, Minh Duc Chu, Kristina Lerman, Emilio Ferrara
TL;DR
This paper presents EDTok, a public dataset of 43,040 TikTok videos related to eating disorders collected from 2019 to 2024 using ED-related keywords and hashtags. It combines video metadata, transcripts, and large-scale user engagement data with comments, enabling multimodal analyses. A two-step filtering pipeline—manual relevance checks and Google Gemini classification—ensures content relevance, while BERTopic and emotion analysis uncover thematic and emotional patterns in descriptions and comments. The dataset supports research on content diffusion, moderation effectiveness, and the impact of the pandemic on eating-disorder discourse, with implications for digital health interventions and platform policies.
Abstract
Eating disorders, which include anorexia nervosa and bulimia nervosa, have been exacerbated by the COVID-19 pandemic, with increased diagnoses linked to heightened exposure to idealized body images online. TikTok, a platform with over a billion predominantly adolescent users, has become a key space where eating disorder content is shared, raising concerns about its impact on vulnerable populations. In response, we present a curated dataset of 43,040 TikTok videos, collected using keywords and hashtags related to eating disorders. Spanning from January 2019 to June 2024, this dataset, offers a comprehensive view of eating disorder-related content on TikTok. Our dataset has the potential to address significant research gaps, enabling analysis of content spread and moderation, user engagement, and the pandemic's influence on eating disorder trends. This work aims to inform strategies for mitigating risks associated with harmful content, contributing valuable insights to the study of digital health and social media's role in shaping mental health.