Polarization in social media assists influencers to become more influential: analysis and two inoculation strategies

Iván Garibay, Alexander V. Mantzaris, Amirarsalan Rajabi, Cameron E. TaylorScientific 2019 · DOI

Abstract

This work explores simulations of polarized discussions from a general and theoretical premise, specifically investigating whether subgroups in online social networks find disagreement beneficial. Using a methodological framework based on stochastic message exchange, the study models key factors in social media engagement, including iterative influence accumulation and asymmetric message treatment during disagreements. The results demonstrate that while influence naturally trends toward a uniform distribution in non-polarized environments, a polarization event reverses this trend, favoring high-influence nodes. The paper discusses the real-world analogues of this reversal and proposes two topic-agnostic, privacy-preserving inoculation strategies to restore influence equality.

Introduction

In a healthy, non-polarized social network, influence tends to become more distributed over time. As users interact, lower-ranked accounts can "piggy-back" on the reach of larger ones, gradually narrowing the gap between the "elite" and the "average" user. This natural democratization of influence is a hallmark of open communication.

However, social media polarization disrupts this process. When a community splits into opposing factions, the resulting "intimidation" and "mixed message" effects disproportionately silence or neutralize the influence of lower-ranked users. This creates a power vacuum that high-influence accounts fill to maintain their dominance. This paper seeks to model this incentive structure and find ways to mitigate it without resorting to censorship or invasive content monitoring.

Methodology

The researchers utilized a methodological framework based on stochastic message exchange equations, extending the "Dynamic Communicators" model.

  • Influence Accumulation: Each node (user) has an influence parameter ($l_n$). When a user’s message is successfully shared by another, their influence score increases.
  • The Polarization Model: The network is divided into two groups. During a polarization event, the probability of sharing content is modified: users favor their own group and ignore content from the opposing group.
  • Simulation Phases: The authors ran simulations through three distinct phases: non-polarized (normal interaction), polarized (disagreement introduced), and inoculation (applying mitigation strategies).

Crucially, the model is topic-agnostic. It does not analyze content sentiment, only the mechanics of who is sharing and their relative influence rankings.

The Reversal of Influence Equality

A key insight of the study is that polarization serves as a structural incentive for top-tier influencers. In a non-polarized state, the ratio of influence between the top and bottom halves of the network steadily decreases, trending toward equality. Once polarization is introduced, this trend is abruptly reversed.

The ratio of responses between the top and bottom halves of the network decreases until polarization is introduced (at time step 30,000), at which point the trend reverses.

As shown in the figure above, the introduction of polarization causes the response ratio to spike. This indicates that the "elite" users are regaining their disproportionate power at the expense of the "long tail" of smaller users. In a polarized state, messages from low-influence users on the opposing side are easily ignored or canceled out. High-influence users, however, possess enough "klout" to punch through the noise, effectively silencing the competition from lower tiers.

Effective responses relative to the top node (Node 1) before and after polarization, showing the disproportionate suppression of lower-ranked nodes.

The comparison of effective responses relative to the top node (Node 1) illustrates this suppression. Post-polarization, the relative influence of lower-indexed nodes is significantly curtailed compared to the pre-polarization state, where influence was more evenly distributed.

Inoculation Strategies

The paper proposes two privacy-preserving strategies to restore influence equality without resorting to censorship or user removal:

  1. Neutralization: This strategy encourages lower-ranked nodes to "opt-out" of the polarized exchange. By neutralizing their content treatment, the network can unilaterally decrease the influence disparity.
  2. Tiered Communication: This strategy segments users into tiers of similar influence, where users primarily see content from others in their own "weight class." This was shown to be highly effective, rapidly returning the network to a uniform distribution of dissemination capabilities by removing the "intimidation" of high-influence accounts.

Conclusion

This research highlights that polarization is not merely a social byproduct but a mechanism that benefits top-tier influencers by reversing the natural trend toward influence equality. By understanding these structural incentives, platforms can implement topic-agnostic inoculation strategies—such as tiered communication—that promote a more equitable distribution of influence while preserving user privacy and platform competition.

Summary updated

Abstract

This work explores simulations of polarized discussions from a general and theoretical premise. Specifically the question of whether a plausible avenue exists for a subgroup in an online social network to find a disagreement beneficial and what that benefit could be. A methodological framework is proposed which represents key factors that drives social media engagement including the iterative accumulation of influence and the dynamics for the asymmetric treatment of messages during a disagreement. It is shown that prior to a polarization event a trend towards a more uniform distribution of relative influence is achieved which is then reversed by the polarization event. The reasons for this reversal are discussed and how it has a plausible analogue in real world systems. A pair of inoculation strategies are proposed which aim at returning the trend towards uniform influence across users while refraining from violating user privacy (by remaining topic agnostic) and from user removal operations.
BibTeX
@inproceedings{garibay2019polarization,
  author = {Iván Garibay and Alexander V. Mantzaris and Amirarsalan Rajabi and Cameron E. Taylor},
  title = {{Polarization in social media assists influencers to become more influential: analysis and two inoculation strategies}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-55178-8},
}
Figure