The Invisible Political Officer: How Personalization Algorithms Shape Public Opinion

Abstract

Social media have been firmly entrenched in the modern everyday life. Still, their influence on the formation of public opinion is not well understood. An important feature of social media is that they are not neutral. Not only do people interact with each other on social media platforms, but social media themselves actively interact with people, selecting personalized content for them based on the information about their interests and behavior. In 2011, Eli Pariser hypothesized that content personalization should lead to the formation of a kind of “information cocoons”, or “filter bubbles” — homogeneous groups of users who hold similar views. However, the fragmentation of the Internet community into “filter bubbles” is not the only threat posed by the use of personalization algorithms. Even more dangerously, social media possess the ability to manipulate content selection algorithms in order to influence users’ views. The article attempts to test the reality of these threats through computational modeling. To solve this task, the author employs a simple agent-based model that simulates the impact of personalization algorithms on communication in social media. The article demonstrates that, contrary to Pariser’s hypothesis, algorithms that select content as close as possible to user preferences result in the emergence of “filter bubbles” rather rarely. The author also finds that manipulation of personalization algorithms makes it possible to influence the formation of public opinion on a stable basis only under two conditions: (1) when all users are manipulated and at the same time they are open to external influence; (2) when manipulation aims at the so called “centrists” who do not possess a clear-cut opinion on some issue.