Opinion Dynamics in Populations of Converging and Polarizing Agents

Opinions determine individuals' attitudes and fundamentally influence collective decisions in societies. As a result, understanding the processes leading to the dynamic formation of opinions is a key research topic across multiple disciplines. Opinion dynamics has been simulated through several computational models where homogeneous agents are assumed to interact over networks. Often, models assume that agents with opposing viewpoints converge in opinion when interacting with each other. This is at odds with evidence showing that individuals can also become further polarized when interacting with individuals having opposing viewpoints. In this paper, we study an opinion dynamics model where both converging and polarizing nodes co-exist in a population. Through simulations on several graph families we aim at understanding i) how radicalization depends on different combinations of such type of nodes and ii) how placing polarizing/converging agents in specific network locations impacts opinion radicalization. We observe that there is an optimal fraction of polarizing agents that minimizes radicalization. Furthermore, we observe that placing polarizing nodes on specific network positions can strongly affect radicalization: assigning high-degree nodes as polarizing results in lower radicalization as compared to random assignment. Our results indicate that considering heterogeneous agents in what concerns their reaction to opposing viewpoints is fundamental to fully grasp the role of social networks in sustaining radical opinions.