New two-layer hypergraph model traces how peer pressure drives opinion spread

Researchers have introduced a new mathematical model that captures how opinions spread through overlapping social settings such as households and workplaces, and have tested which machine learning methods can best recover the model’s parameters from incomplete observations.

The model, developed by Ágnes Backhausz, Villő Csiszár, Csegő Balázs Kolok, Damján Tárkányi and András Zempléni, represents social structure as a two-layer random hypergraph. In this framework, hyperedges stand for individual households and workplaces, allowing interactions that involve more than two people at once. This higher-order structure is important because peer pressure often operates through groups rather than through simple pairwise contacts, and the two layers overlap since the same individuals belong to both a household and a workplace.

Within this adaptive structure, individuals can react when their opinion is in the minority in their groups. The process unfolds through stochastic steps: a person may either change their opinion to match the majority, or quit their workplace and join another one in which their opinion already belongs to the majority. This second mechanism, switching groups rather than changing minds, is what makes the model adaptive, since the network itself is reshaped by the dynamics taking place on it.

The authors pursued two main goals. First, using computer simulations, they described how the parameters governing opinion change and workplace switching affect three observable outcomes: homophily, the speed of polarization, and the number of components in the network. Second, they analyzed the model as a Markov chain and studied its absorbing states, the configurations in which the process can no longer change. This theoretical grounding complements the simulation-based picture of how the system behaves under different parameter settings.

The third strand of the study is practical: estimating the model’s parameters from partial observations of the process, such as the distribution of opinion configurations within households and workplaces. The team quantitatively compared how four methods perform on this task: linear regression, XGBoost, a convolutional neural network (CNN) and a long short-term memory network (LSTM). Each method was asked to recover the underlying parameters from limited data, mimicking the realistic situation in which an observer cannot track every individual interaction.

The results showed a clear pattern. In most cases, especially for shorter observation periods, the LSTM performed best, while in many cases XGBoost was also a strong contender. For longer time horizons, the CNN became competitive as well.

One particularly important observation concerned the data itself: the amount and kind of information required for accurate parameter estimation depends on the strength of the peer pressure effect.

The authors note that the work is based on simulations and theoretical analysis of the model itself, so the findings concern the behavior of the proposed model rather than measurements from a specific real-world population. The framework is open access and available to researchers studying higher-order networks, opinion formation and parameter estimation from partial observations.

Subject of Research: Adaptive two-layer hypergraph modeling of opinion spread with parameter estimation by machine learning methods

Article Title: A new adaptive two-layer model for opinion spread in hypergraphs: parameter sensitivity and estimation

Article References: Backhausz, Á., Csiszár, V., Kolok, C. B., Tárkányi, D., & Zempléni, A. (2026). A new adaptive two-layer model for opinion spread in hypergraphs: parameter sensitivity and estimation. Applied Network Science. https://doi.org/10.1007/s41109-026-00829-9

Image Credits: AI Generated

DOI: 10.1007/s41109-026-00829-9

Keywords: opinion dynamics, hypergraphs, peer pressure, Markov chains, LSTM, XGBoost, convolutional neural networks, parameter estimation, homophily, polarization, network science, stochastic modeling