Quantifying and Predicting Mental Illness Severity in Online Pro-Eating Disorder Communities

Chancellor, Stevie; Lin, Zhiyuan; Goodman, Erica L.; Zerwas, Stephanie; Choudhury, Munmun De

doi:10.1145/2818048.2819973

Cited by 153 publications

(126 citation statements)

References 42 publications

Supporting

Mentioning

123

Contrasting

Order By: Relevance

“…As shown in prior work [14], humans are good at identifying the obvious extremes of what abides by community guidelines (e.g., fitness posts versus pro-ED diet advice). However, certain pro-ED posts are challenging for humans and moderators alike (e.g., thinspiration versus pro-ED diet advice).…”

Section: Methodological Challenges In Detecting Deviant Contentmentioning

confidence: 90%

“…However, due to the inherent subtleties in pro-ED posts, there is a major resource cost as well as an emotional cost to identifying sensitive material [66]. It may also be challenging to find crowdworkers with significant skill to accurately assess these posts [14].…”

Section: Methodological Challenges In Detecting Deviant Contentmentioning

confidence: 99%

“…Because of this, the content around eating disorder communities online is contextually complex. On Instagram, users can move between promoting eating disorders and struggling to stop their behaviors on the same user account [14]. These nuances have led platforms like Tumblr to establish community guidelines that distinguish between these kinds of content [85].…”

Section: Pro-ed Communitiesmentioning

confidence: 99%

“…Pro-ED communities have challenged social media platforms because of their pervasiveness despite explicit moderation against the community [17]. Not all content in pro-ED communities is dangerous [14] nor do all images qualify for removal-the same photo of a model could be used for a pro-ED blog or a fitness blog for motivation. However, for the content that is dangerous, psychological research shows that there are unique social contagion-like effects on those who are exposed to this content [10].…”

Section: Introductionmentioning

confidence: 99%

“…Prior work in HCI, mental wellness, and prediction has focused on the identification of specific mental wellness challenges like depression [26] or assessing more general wellness states of those who have posted to pro-ED communities [14]. This work is distinct from these works in three major ways.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multimodal Classification of Moderated Online Pro-Eating Disorder Content

Chancellor

Kalantidis

Pater

et al. 2017

Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems

Self Cite

View full text Add to dashboard Cite

Social media sites are challenged by both the scale and variety of deviant behavior online. While algorithms can detect spam and obscenity, behaviors that break community guidelines on some sites are difficult because they have multimodal subtleties (images and/or text). Identifying these posts is often regulated to a few moderators. In this paper, we develop a deep learning classifier that jointly models textual and visual characteristics of pro-eating disorder content that violates community guidelines. Using a million Tumblr photo posts, our classifier discovers deviant content efficiently while also maintaining high recall (85%). Our approach uses human sensitivity throughout to guide the creation, curation, and understanding of this approach to challenging, deviant content. We discuss how automation might impact community moderation, and the ethical and social obligations of this area.

show abstract

Section: Methodological Challenges In Detecting Deviant Contentmentioning

confidence: 90%

Section: Methodological Challenges In Detecting Deviant Contentmentioning

confidence: 99%

Section: Pro-ed Communitiesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Multimodal Classification of Moderated Online Pro-Eating Disorder Content

Chancellor

Kalantidis

Pater

et al. 2017

Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems

Self Cite

View full text Add to dashboard Cite

show abstract

Assessment of Accuracy, User Engagement, and Themes of Eating Disorder Content in Social Media Short Videos

2023

View full text Add to dashboard Cite

ImportanceSocial media is simultaneously home to communities of users who promote eating disorders as a lifestyle and users who advocate for recovery. As studies have confirmed an association between exposure to pro–eating disorder content and engaging in disordered eating behaviors, an examination of the accuracy of and interactions with information shared in these complex and contradictory communities can provide insights into content available to users at risk.ObjectiveTo determine the associations among themes, accuracy of information, and user engagement of eating disorder content on a short video–sharing social media platform.Design, Setting, and ParticipantsThis qualitative study included a thematic analysis of a sample of 200 TikTok videos, as well as user engagement metrics and content creator characteristics, between February and June 2022. Data were analyzed from March to June 2022.Main Outcomes and MeasuresContent themes, accuracy of information, user engagement, and the associations among these factors were identified in a sample of eating disorder videos on a social media platform. Data were analyzed with Pearson χ2, analysis of variance, linear regression, and random permutation tests.ResultsAmong the 200 videos assessed, 124 (62.0%) covered prorecovery content, 59 (29.5%) included pro–eating disorder content, and 17 (8.5%) contained anti–eating disorder content. Thematic analysis revealed 4 high-level themes: (1) encouraging the development or sustainment of eating disorders, (2) sharing physical or emotional experiences with eating disorders, (3) sharing narratives of recovery, and (4) social support. While the results of Pearson χ2 test indicated that videos in the prorecovery domain contained more accurate content than videos categorized in the pro–eating disorder and anti–eating disorder domains (χ24 = 157.92; P &lt; .001), the results of analysis of variance demonstrated that there was no significant difference in user engagement between informative and misleading content (likes: F = 0.110; P = .95; comments: F = 2.031; P = .13; views: F = 0.534; P = .59; shares: F = 0.691; P = .50). Results of the 10 000 random permutation tests, in which all P values were between .40 and .60 regardless of the distances, suggested that there was no significant difference in user engagement among the 3 domains.Conclusions and RelevanceThis mixed-methods qualitative analysis of misleading eating disorder information on social media found that pro–eating disorder and prorecovery communities were prevalent. However, social media users in the prorecovery community created more informative than misleading content. Furthermore, the study found no significant difference between users’ engagement with accurate vs misleading videos, which may suggest that false information alone does not influence the virality of videos.

show abstract