Dynamic Parameter Calibration Framework for Opinion Dynamics Models

Abstract

In the past decade, various opinion dynamics models have been built to depict the evolutionary mechanism of opinions and use them to predict trends in public opinion. However, model-based predictions alone cannot eliminate the deviation caused by unforeseeable external factors, nor can they reduce the impact of the accumulated random error over time. To solve this problem, we propose a dynamic framework that combines a genetic algorithm and a particle filter algorithm to dynamically calibrate the parameters of the opinion dynamics model. First, we design a fitness function in accordance with public opinion and search for a set of model parameters that best match the initial observation. Second, with successive observations, we tracked the state of the opinion dynamic system by the average distribution of particles. We tested the framework by using several typical opinion dynamics models. The results demonstrate that the proposed method can dynamically calibrate the parameters of the opinion dynamics model to predict public opinion more accurately.

Bounded Confidence and Cohesion-Moderated Pressure: A General Model for the Large-Scale Dynamics of Ordered Opinion

“…As for (2), its fundamental concept revolves around measuring the “distance” between simulated data and real data, thereby identifying the optimal parameter values that sufficiently minimize this discrepancy through appropriate optimization methods. This approach frequently relies on intelligent algorithms and stochastic optimization techniques, such as particle swarm optimization or genetic algorithms [ 39 , 40 , 41 ], which offer reliable search strategies for “finding parameters that align with reality”. The convergence of these algorithms has been effectively proven under certain conditions.…”