Abstract
We introduce multi-population opinion dynamics models linked to the bounded confidence model, aiming to explore how interactions between individuals contribute to the emergence of consensus, polarization, or fragmentation. Existing models either neglect agent similarities, sacrificing accuracy for scalability, or prioritize accuracy by introducing agent-wise connections, constraining scalability. Our proposed model captures similarities between agents in a scalable manner. In our setting, agent similarities are defined by their group affiliations. Specifically, each sub-population is characterized by its distribution, and the closeness between two sub-populations is measured by the Wasserstein distance of their corresponding distributions. This leads to two mutually connected dynamics: micro, the individual-based dynamics, and macro, the distribution-based one. The individual-wise interactions take into account the population-wise interactions (similarities), and the population-wise interactions are updated based on the individual-wise interactions. We have proven the well-posedness of our models. Additionally, we conducted several simulations to mimic certain complex social phenomena.
Introduction
The evolution of opinions in large social networks is a complex process driven by individual interactions and broader group identities. Traditional "Bounded Confidence" (BC) models, which posit that individuals only interact with those holding similar views, have been foundational in the field. However, these models often struggle with a fundamental trade-off: simple models that treat all agents as identical are computationally efficient but lack social nuance, while detailed models tracking individual connections are difficult to scale to millions of users.
This paper addresses this gap by introducing a multi-population framework that incorporates group identity into the mathematical dynamics. By modeling groups as distributions and using the Wasserstein distance to measure similarity, we create a scalable model that captures the interplay between individual opinions and collective affiliations.
Methodology
The core of our approach lies in the coupling of micro-level individual dynamics with macro-level population distributions.
Local Kernel Methods
To model the influence between agents, we move beyond simple binary thresholds and utilize local kernel functions. These kernels define how much one individual influences another based on the distance between their opinions.
As illustrated in the figure above, the choice of kernel—Uniform, Exponential, or Triangular—significantly affects the convergence of opinions. While Uniform and Exponential kernels often lead to consensus under certain parameters, the Triangular kernel can naturally lead to polarization into distinct clusters.
Wasserstein Distance and Group Identity
A significant innovation in this model is the use of the Wasserstein distance (or "Earth Mover's Distance") to quantify the similarity between entire sub-populations. Instead of reducing a group to a single average opinion, we treat each group as a probability distribution. This allows the model to account for internal diversity and the influence of "fringe" members who may bridge the gap between disparate groups.
Micro-Macro Coupling
The model links individual opinion changes (micro) to shifts in the overall group distribution (macro). Individuals update their views based on their peers and the perceived similarity (or bias) between their own group and others. For extremely large populations, we derive the mean-field limit, transitioning from individual ordinary differential equations (ODEs) to a system of continuity equations (PDEs), which allows for the mathematical description of massive populations.
Results and Discussion
We proved the mathematical well-posedness of the model, ensuring stable and unique solutions. Simulations were then used to explore various social phenomena.
Sensitivity to Influence Parameters
The shape of the influence kernel is a critical factor in determining whether a society reaches consensus or fragments.
The figure above demonstrates the effect of varying the parameter gamma in an exponential kernel. Lower values of gamma (broader influence) promote consensus, whereas higher values (narrower, more "stubborn" influence) lead to increasing levels of fragmentation and the formation of multiple isolated opinion clusters.
Social Phenomena and Polarization
The model successfully reproduced several complex behaviors:
- In-Group Favoritism: Increased group bias leads directly to the formation of echo chambers where opinions converge within a group but diverge from the rest of society.
- Asymmetric Bias: If one group is more receptive than another, the less receptive (more "stubborn") group's opinion tends to dominate the overall narrative, pulling others toward them.
- State-Dependent Resistance: The model accounts for the fact that individuals with extreme opinions are often more resistant to change, providing a mathematical basis for the persistence of polarized viewpoints and the difficulty of breaking echo chambers.
- Polarization Reduction: By introducing a "stubborn neutral" group that acts as an information bridge, we demonstrated that it is possible to reduce polarization between opposing factions.
Conclusion
This multi-population model provides a robust framework for understanding opinion dynamics by bridging the gap between individual interactions and group identities. While the model currently focuses on population distributions, future work will involve validating these findings against real-world social media data and incorporating more complex network topologies. Nonetheless, the integration of Wasserstein distance and micro-macro coupling offers a powerful tool for simulating and potentially mitigating social polarization.
Summary updated