Political communication on social media: A tale of hyperactive users and bias in recommender systems

Abstract

A segment of the political discussions on Online Social Networks (OSNs) is shaped by hyperactive users. These are users that are over-proportionally active in relation to the mean. By applying a geometric topic modeling algorithm (GTM) on German users’ political comments and parties’ posts and by analyzing commenting and liking activities, we quantitatively demonstrate that hyperactive users have a significant role in the political discourse: They become opinion leaders, as well as having an agenda-setting effect, thus creating an alternate picture of public opinion. We also show that hyperactive users strongly influence specific types of recommender systems. By training collaborative filtering and deep learning recommendation algorithms on simulated political networks, we illustrate that models provide different suggestions to users, when accounting for or ignoring hyperactive behavior both in the input dataset and in the methodology applied. We attack the trained models with adversarial examples by strategically placing hyperactive users in the network and manipulating the recommender systems’ results. Given that recommender systems are used by all major social networks, that they come with a social influence bias, and given that OSNs are not per se designed to foster political discussions, we discuss the implications for the political discourse and the danger of algorithmic manipulation of political communication.