Is There Personalization in Twitter Search? A Study on polarized opinions about the Brazilian Welfare Reform

Abstract

Personalization algorithms play an essential role in the way search platforms fetch results to users. While there are many empirical studies about the effects of these algorithms on Web searches like Google and Bing, reports about personalization on social media searches are rare. This exploratory study aims to understand and quantify the limits of personalization in Twitter search results. We developed a measurement methodology and agents to train a pair of polarized Twitter accounts and simultaneously collected search results from these accounts. The agents were run in a political context, the Brazilian Welfare Reform. Our findings show a significant amount of personalization differences when we compare search results from a new fresh profile to non-fresh ones. Peculiarly, little evidence for differences between two profiles that followed different accounts with polarized viewpoints about the same topic was found – the filter bubble hypothesis cannot be null.