Poster: Information Flow Experiments to study News Personalization

There are concerns over online personalization. Through content surveys, Thurman et al. [2] uncover increasing degrees of personalization at top national news websites over a period of three years. Pariser [3] points out that such personalization may trap users in their own customized ‘filter bubble’. Such concerns have led to much interest in determining whether such filter bubbles exist. Englehardt et al. study several news platforms, but do not find high levels of personalization [4]. To detect personalization, we look at the methodology due to Tschantz et al. [5], which demonstrates how to conduct experiments in a systematic way to detect instances of information flow in web systems. Datta et al. [6] extend the methodology due to Tschantz et al. and study properties of transparency, choice, and discrimination on Google Ads and Ad Settings. They developed AdFisher, an automated tool to run browserbased experiments and analyze data using machine learning and significance tests.