Abstract
Users are increasingly exposed to security, privacy, and safety harms via content distributed by platforms, such as ads and social media posts.
Protecting users requires holding platforms accountable, by pressuring trust and safety teams and regulators. But this is challenging because content is ephemeral and targeted, and platforms rarely make such data available to outside researchers. In this talk, I will discuss how mixed-methods measurement techniques, which combine large-scale data collection with qualitative analysis, can be used to collect empirical data on harmful content served by platforms, and how this data can translate into impact. I will showcase this approach through two case studies of problematic online ads. First, I found that a specific set of ad platforms is responsible for most of the deceptive advertising on news websites.
Second, I found that ad platforms on the web perform limited targeting of users based on their health conditions, but nearly half of health advertising is potentially deceptive. I will also discuss how I am extending this approach to other problem areas on online platforms, such as algorithmic social media feeds and non-consensual deepfakes.
Biography
Eric Zeng is a Postdoctoral Fellow at Georgetown University. He received his PhD in Computer Science & Engineering from the University of
Washington. His research empirically measures the safety, security, and privacy risks that users encounter on online platforms. His current interests include Al-generated non-consensual intimate imagery, algorithmic social media feeds, and online advertising.