We’re excited to restart our Internal Team Seminars with Salim, who will present his PhD thesis: “Computational Methods for Analyzing Risks with Information Exposure on Social Media.”
The seminar is held on 17th of February, from 3 to 4:00PM, in Grace Hopper conference room. The presentation will last approximately 45 minutes, followed by a feedback and Q&A session.
We look forward to an engaging discussion—see you there!
Abstract:
With platforms like Facebook now serving as primary news sources, users engage with news content as part of their social interactions while browsing, often without actively seeking it. This introduces several risks, such as political polarization, misinformation, and coordinated influence campaigns. Studying these dynamics is challenging due to limited data access. Researchers face three key obstacles: (1) a lack of comprehensive lists of news producers active on social media, (2) no direct access to the news users see or interact with, and (3) while there is ongoing debates about regulating online advertising to mitigate the risks, there is insufficient understanding of how restrictions on micro-targeting and web tracking might impact the ecosystem.
This thesis addresses these gaps by introducing methodologies, datasets, and analyses, with a focus on Facebook. First, we propose a method using the GNews API and CrowdTangle to identify self-proclaimed news producers on social media. This approach identified over 26,000 U.S.-based news sources—far more than the 4,323 sources listed by Media Bias Fact Check and News Guard, used in previous research. Next, we introduce a privacy-preserving tool that captures user interactions with Facebook news. Analyzing data from 472 U.S.-based participants, we find that users encounter misinformation primarily due to their own selection of low-quality sources, rather than exposure via friends or platform algorithms. Moreover, users engage with politically opposing sources when such interactions remain private from their social circles.
We then assess the role of micro-targeting techniques in online advertising, showing that restricting advertiser-driven techniques may have little impact on how advertisers use the platform and may not sufficiently mitigate the risks. However, we highlight another form of targeting—algorithmic-driven micro-targeting—where Facebook assigns a relevancy score to optimize ad distribution. We argue that regulations should address this as well. Lastly, we explore the ads.txt standard as a tool for detecting coordinated influence campaigns. We develop a clustering method that groups websites based on ads.txt similarities, suggesting potential shared ownership. While further validation is required, this method provides a foundation for identifying networks of affiliated news sites.
