Doppelg\"anger Model: Emergence of polarization in opinion dynamics

TL;DR

This paper introduces a novel opinion dynamics model incorporating negative interactions as attraction to a Doppelgänger, revealing four regimes including polarization that mirrors real-world online discourse structures.

Contribution

The study presents a new network-based opinion formation model with a unique negative interaction dynamic, capturing polarization and echo chamber phenomena.

Findings

Model exhibits four distinct regimes of opinion dynamics.

Polarization regime replicates empirical online discourse structures.

Rich behaviors emerge from the novel negative interaction dynamic.

Abstract

Over the past decade, contrary to the early popular expectation that large-scale discourse in online communities would foster greater consensus, the large-scale structure of online discourse has been measured to be strongly polarized. Though it was posited that this effect was driven mainly by the algorithmic curation of content by social platforms (the "Filter Bubble") it appears that polarization is mainly driven by the tendency of interactions among users to be homophilic; users strongly prefer to endorse content and other users that are similar to them. Users organize into homophilic clusters or "echo chambers" where interactions within clusters are more likely to be positive and interactions between clusters are more likely to be negative. This motivates studying polarization in online discourse through modeling as an example of self organized dynamics. Models of opinion formation on networks have tended to only encode positive interactions (attraction) between users - as a consequence consensus or non-interacting clusters of users are ubiquitous behaviors. In this research a model of network-based opinion formation is introduced that includes a novel dynamic of negative interaction; instead of modeling negative interaction as repulsion in the opinion space it is modeled as an attraction to a "Doppelg\"anger" user. It is demonstrated that this model exhibits rich behavior. Four distinct regimes of local interaction correspond to different population-wide dynamics - Neutral Consensus, Extreme Consensus, Polarization and Undecided. In the Polarization regime, local interactions are modeled to be homophilic and the structure of discourse that emerges replicates what is seen in empirical data; clustered users where interactions within clusters is positive and interaction between clusters is negative.