Doppelgänger Model: Emergence of polarization in opinion dynamics

Vince Campo, Sebastien Motsch, Dylan Weber · 2025

Read Doppelgänger Model: Emergence of polarization in opinion dynamics by Vince Campo, Sebastien Motsch, Dylan Weber (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-14900.txt.

Introduction

Vince Campo, Sebastien Motsch, and Dylan Weber model how online opinion splits into camps, in this 2025 paper. Many models let users only attract each other, which yields consensus or silent clusters. Their model adds a twist for negative contact: instead of pushing away in opinion space, a user is drawn toward a mirror double, a doppelgänger.

Four regimes appear as settings change: neutral consensus, extreme consensus, polarization, and an undecided state. In the polarization regime, the model mimics real discourse, with warm ties inside clusters and cold ties across them. The authors offer it as a self-organization account of echo chambers.

Catalog record

CREATORVince Campo, Sebastien Motsch, Dylan Weber
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-14900

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Doppelgänger Model: Emergence of polarization in opinion dynamics Vince Campo, Sebastien Motsch, Dylan Weber 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 “Doppelggänger" 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. 1 Introduction The internet was and continues to be lauded by many for its role in improving discourse and democratizing information by allowing for more direct discussion among the populace and greatly enhancing access to information generally [ 45 , 3 , 23 , 43 ] . However at its advent and throughout its history some have cautioned that due to well established psychological tendencies (like confirmation bias and selective exposure), discourse on the internet would actually result in a narrowing of the information landscape of its users. This view predicted, given the freedom to select information from the virtually infinite array that the internet provides, that users would spend their time on the internet interacting with those who they already agreed with and consuming information from sources that confirmed their prior views. In other words this view predicted that the structure of online discourse would tend towards "echo chambers" or homophily [ 21 , 20 , 13 , 42 , 30 , 51 , 50 ] . The 2016 United States presidential election prompted many empirical studies leveraging large social media datasets looking into the extent to which homophily presents on social media and what effects it might have on its users. Despite the large variety of methods across this body of work, it is largely unified in its findings; social media has a strongly homophilic structure (see, for example, Figure 1 ) and this structure contributes to multiple undesirable effects including the narrowing of user information landscapes, enhanced spread of misinformation in homophilic clusters and increased ideological polarization [ 6 , 7 , 25 , 26 , 17 , 38 , 47 , 4 , 12 , 2 , 44 , 58 , 10 , 48 , 27 , 18 , 16 , 1 , 22 , 53 ] . Many point to the algorithmic curation of content by social platforms as a main culprit, often referred to as the “Filter Bubble". This certainly has an effect however, it seems that our own tendency to self-select ideologically aligend content is a more important driver . This motivates the study of the dynamics of online discourse through modeling as an example of self organized dynamics ; large scale behaviors emerge without a central authority much like the behavior of a flock of birds or shoal of fish [ 14 , 33 , 15 , 5 , 19 , 52 ] . In this arena, models are usually posed in an agent based framework where the "opinion" of an agent is modeled as a continuous value and is influenced, according to a model of local interactions, by those to whom it is connected in a directed network. Of particular interest is how the topology of the underlying network affects the observed distribution of opinions and vice versa [ 37 , 46 , 55 , 32 ] . Often, the interaction models carry an assumption of local consensus ; if agents interact only with each other than they should agree in some sense. This assumption can also be interpreted as asserting that there are only attractive forces (equivalently, only positive weights in the interaction network) present among the opinions of the agents. Additionally, the underlying network topology is often assumed to be homophilic; agents only interact if their opinions are within some threshold of each other. These assumptions cause the formation of static clusters of agents or a consensus in the long time limit to be a generic behavior [ 28 , 41 , 57 , 35 , 15 , 49 , 11 , 9 ] . While the interaction networks generated by these types of assumptions have been seen to replicate some of the features of interaction patterns seen in real social media data [ 48 , 8 , 22 ] , they usually do not allow for the existence of a repulsive interaction (equivalently, negative weights in the interaction network) between agents. This is problematic from the modeling point of view as it is well established that a negative interaction can result in a backlash effect where both individuals become further polarized from each other [ 34 , 58 , 31 , 36 , 39 , 40 ] . Additionally, under local consensus assumptions, homophily in the interaction network presents as disconnected network components who do not interact, whereas there is evidence that in social media there is significant negative interaction between homophilic clusters [ 56 , 54 , 24 ] . Figure 1 : Visualization of the Twitter conversation surrounding the Black Lives Matter movement from May 2023. Nodes in the network are users, edges represent aggregated interactions between users over the data collection period. Nodes are colored according to their support for or opposition to the Black Lives Matter movement - red indicates support, blue indicates opposition. Edges are directed and are colored according to the color of the source node. The network illustrates a strongly homophilic structure. In this paper we introduce and study a class of models that include a mechanism for negative interaction. Mathematically, inclusion of negative interactions can be problematic; the obvious choice to model them is a repulsion between the interacting agents in the opinion space which can cause the model to diverge. In this work we introduce a dynamic to address this difficulty; when one agent (the “interactor") directs a negative interaction at another agent (the “interactee"), instead of the interactee feeling a “push" on their opinion away from the interactor’s, they feel a “pull" towards the opinion that

Read the complete text