Stevie Chancellor
Georgia Institute of Technology
schancellor3@gatech.edu
Bio
Stevie Chancellor is a PhD candidate in Human Centered Computing in Interactive Computing at Georgia Tech. She is advised by Munmun De Choudhury. Her research interest lies in quantitative social computing and human computer interaction. Specifically she studies computational methods to understand deviant mental health behavior in online communities and combines techniques from Natural Language Processing Machine Learning and Data Science. Her work has won several Best Paper Honorable Mention awards at CHI and CSCW premier venues in human computer interaction. Prior to GT Stevie received a BA from the University of Virginia in 2012 and an MA from Georgetown University in 2014. Her work is supported by a Snap Inc. Research Fellowship and has appeared in national publications such as Wired and Gizmodo.
Computational Methods to Understand Deviant Health Behaviors in Online Communities
Computational Methods to Understand Deviant Health Behaviors in Online Communities
Online groups provide advice and a sense of community for health and wellness, especially for stigmatized mental health disorders. However, individuals also turn to online communities to promote deliberate self-injury disordered eating habits and suicidal ideation. These behaviors can have contagion-like effects on those outside of these communities as well as impacts on social networks that struggle with managing such dangerous content. These behaviors are “deviant” – actions that violate the social norms and behaviors of a particular community. Deviant behaviors that promote self-injury violate platform policies as well as social expectations that individuals do not harm themselves. To precisely identify and manage these pernicious issues is a challenge by itself as is designing appropriate interventions. I study deviant behavior in online communities using computational techniques on large-scale social data. My area of focus is deviant mental health behaviors such as pro-eating disorder. I draw from machine learning and computational linguistics to analyze social media datasets to understand patterns of behavior. Taking a human-centered approach to this complex topic, I explore issues of content moderation/management and ethics in managing these behaviors. I integrate these computationally robust techniques with human knowledge and domain expertise alongside ethics to design “human-in-the-loop” systems for answering our toughest questions about deviant behavior online. My dissertation research focuses on pro-eating disorder communities, a clandestine group that glorifies eating disorders. In my prior work, I analyzed the patterns of elaborate yet semantically meaningful lexical variants to evade content bans on these communities on Instagram such as changing tags from “thighgap” to “thyghgappp”. In another work, I developed a new model of generalized mental illness severity on online data that predicts future severity levels up to 8 months in advance.