A society-wide failure of algorithmic transparency is currently perpetuating innumerable social, economic and health related risks around the world. While algorithms play a role in the proliferation of extremism online — be it white nationalism or other toxic social phenomena — we know very little about how these algorithms operate and how their operation is affecting users; researchers have no robust way to probe how or why extremism emerges and the role that algorithms play in its development. Our team is working on causal investigations of algorithmic responses to user behavior and the evolving dialectic between user behavior and algorithmic recommendations.
Bot detection is an important part of keeping social media platforms safe. State-of-the-art computing tools appear to have high accuracy at this task on benchmark datasets, but this is attributable to limitations in data collection and
labeling strategies. This raises doubts about the effectiveness of the tools social platforms use.
Studies have indicated that ads on social platforms can be delivered to demographic groups at different rates. Few studies explore this by collecting data about what users are seeing. A novel dataset gives researchers a clearer picture of how ads for housing, jobs, and credit proliferate to different racial groups.
What influence do YouTube algorithms have on political thought and radicalization? Researchers take a new approach, mapping content and users in a multi-dimensional space based on more granular information about their consumption than previous studies – in particular, the signals embedded in videos' scripts.
If people are exposed to discriminatory content online, it may potentially reinforce harmful behavior offline. This project seeks to understand whether people’s identities and attributes contribute to the relationship between online consumption and offline behavior.
Chris Hays, Zachary Schutzman, Manish Raghavan, Erin Walk, and Philipp Zimmer. 2023. Simplistic Collection and Labeling Practices Limit the Utility of Benchmark Datasets for Twitter Bot Detection. In Proceedings of the ACM Web Conference 2023 (WWW ’23). Association for Computing Machinery, New York, NY, USA, 3660–3669. https://doi.org/10.1145/3543507.3583214
MIT Sloan’s Ideas Made to Matter highlights the Social Media vertical’s recently published study that shows “general-purpose bot-detection algorithms trained on a particular data set may be highly error-prone when applied in real-world contexts.”
SES student Chris Hays received the award for the group’s publication “Simplistic Collection and Labeling Practices Limit the Utility of Benchmark Datasets for Twitter Bot Detection”.
Co-lead by Sinan Aral (David Austin Professor of Management at the MIT Sloan School of Management) and Dean Eckles (Associate Professor of Marketing at MIT Sloan), the Social Media vertical team consists of Chris Hays (MIT PhD Student, IDSS Social & Engineering Systems), Manish Raghavan (Drew Houston (2005) Career Development Professor at MIT Sloan and MIT EECS), Erin Walk (MIT PhD Student, IDSS Social & Engineering Systems), and Philipp Zimmer (MIT Technology and Policy Program and EECS SM).