Evolution of beliefs in social networks

Pushpi Paranamana, Pei Wang, Patrick ShaftoarXiv 2022 · OpenReview

Abstract

Evolution of beliefs of a society are a product of interactions between people (horizontal transmission) in the society over generations (vertical transmission). Researchers have studied both horizontal and vertical transmission separately. Extending prior work, we propose a new theoretical framework which allows application of tools from Markov chain theory to the analysis of belief evolution via horizontal and vertical transmission. We analyze three cases: static network, randomly changing network, and homophily-based dynamic network. Whereas the former two assume network structure is independent of beliefs, the latter assumes that people tend to communicate with those who have similar beliefs. We prove under general conditions that both static and randomly changing networks converge to a single set of beliefs among all individuals along with the rate of convergence. We prove that homophily-based network structures do not in general converge to a single set of beliefs shared by all and prove lower bounds on the number of different limiting beliefs as a function of initial beliefs. We conclude by discussing implications for prior theories and directions for future work.

Introduction

How do societies arrive at their collective beliefs? Historically, this question has been approached through two distinct lenses. Vertical transmission, often the focus of cognitive science, examines how beliefs—such as language or cultural norms—are passed down through generations, typically modeled as Bayesian agents learning from predecessors. Conversely, horizontal transmission, studied extensively in network science, focuses on the peer-to-peer spread of information, opinions, or diseases within a single generation.

The reality of belief formation involves both forces acting simultaneously: we learn from our parents and our peers. This paper introduces a unified mathematical framework using Markov chain theory to bridge this gap, exploring when a society reaches a homogeneous consensus versus when it fractures into heterogeneous subgroups.

Methodology

The proposed framework represents a society’s beliefs at any time $n$ as a row-stochastic matrix $Q_n$. This matrix is derived from three primary components: the network structure ($P$), the initial priors ($M$), and the concept structure ($H$).

Notation and definitions for the belief evolution framework.

The evolution of beliefs is modeled by the product of these matrices over time. The research analyzes three specific network regimes:

  1. Static Networks: The social structure remains constant over time.
  2. Randomly Changing Networks: The social structure evolves independently of individual beliefs.
  3. Homophily-based Dynamic Networks: The network structure $P$ is updated at each step based on the Kullback-Leibler (KL) divergence between individuals' beliefs. If the "distance" between two individuals' beliefs exceeds a threshold $\epsilon_p$, communication ceases.
Example of a social network structure $P$ with three individuals $\alpha_1, \alpha_2, \alpha_3$.

Key Insights and Results

The Dominance of Concept Structure

A significant finding is that if the concept structure $H$ is indecomposable—meaning any concept can eventually be confused with any other over generations—the society will eventually reach a consensus. This holds true even if the social network is fragmented into isolated groups that never interact.

Homophily and Polarization

Unlike static or random networks, homophily-based networks act as engines for polarization. These structures create "KL clusters" that do not generally converge to a single belief. The authors provide a lower bound on the number of distinct belief groups that emerge based on initial priors. Simulations show that in these models, links are frequently destroyed as beliefs drift apart, leading to the formation of "echo chambers."

Convergence and Heterogeneity

For static and randomly changing networks, the authors proved that if the network is "SIA" (Stochastic, Indecomposable, Aperiodic), the society converges to a state where everyone holds the same belief distribution. The rate of this convergence is mathematically tied to the second-largest eigenvalues of the $P$ and $H$ matrices; the slower process (social vs. cultural) dictates the overall evolution speed. For homophily models, stable heterogeneity is possible, and the paper introduces an algorithm to predict the minimum number of sub-communities formed.

Conclusion and Future Work

While this framework provides a robust theoretical foundation, it currently models closed systems without external information input (e.g., news or scientific discovery). Furthermore, it assumes "honest" actors, leaving the impact of adversarial misinformation for future study. Future research should also incorporate population dynamics (birth and death of agents) and calibrate these models with empirical behavioral data to validate the predicted KL-divergence thresholds in human belief shifts.

Summary updated

Abstract

Evolution of beliefs of a society are a product of interactions between people (horizontal transmission) in the society over generations (vertical transmission). Researchers have studied both horizontal and vertical transmission separately. Extending prior work, we propose a new theoretical framework which allows application of tools from Markov chain theory to the analysis of belief evolution via horizontal and vertical transmission. We analyze three cases: static network, randomly changing network, and homophily-based dynamic network. Whereas the former two assume network structure is independent of beliefs, the latter assumes that people tend to communicate with those who have similar beliefs. We prove under general conditions that both static and randomly changing networks converge to a single set of beliefs among all individuals along with the rate of convergence. We prove that homophily-based network structures do not in general converge to a single set of beliefs shared by all and prove lower bounds on the number of different limiting beliefs as a function of initial beliefs. We conclude by discussing implications for prior theories and directions for future work.
BibTeX
@misc{paranamana2022evolution,
  author = {Pushpi Paranamana and Pei Wang and Patrick Shafto},
  title = {{Evolution of beliefs in social networks}},
  year = {2022},
  eprint = {2205.13587},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.13587},
}
Figure