Control in Social and Economic Systems

Forecasting the dynamics of public opinion based on longitudinal data of high granularity: the Abelson model, regression models, and ensembles of models

M. E. Buzikov^a, I. Petelina^b, S. A. Krasotkin^a, M. S. Ryzhov^a, I. V. Kozitsin<sup>ac

a</sup> V.A. Trapeznikov Institute of Control Sciences of RAS, Moscow
^b Ozon Tech, Moscow
^c Moscow Institute of Physics and Technology, Moscow

Abstract: We consdier the problem of forecasting the dynamics of public opinion based on longitudinal data of high granularity gleaned from the social network VKontakte. This problem was suggested to the participants of the Hackathon "UBS Challenge'2024" as one of the leisure events of the XX All-Russian School-Conference of Young Scientists "Management of Large Systems" (UBS), held in Novocherkassk in 2024. This paper is devoted to a detailed description of the Hackathon and the solutions proposed by its participants. For a sample of N = 1 648 829 users, based on two granular snapshots of their opinions taken six months apart (in February and July 2018), participants have to elaborate on a forecast of the distribution of public opinion in December 2018. The participants also had the information about the structure of friendship ties between users. We report that the highest accuracy was achieved by an ensemble of two models – the Abelson model, enhanced by estimating users' social power via the eigenvector centrality measure, and the constant trend model.

Keywords: social influence models, artificial intelligence, social networks, eigenvector centrality, ensembles of models.

UDC: 519.7
BBK: 22.18

Received: January 28, 2025
Published: May 31, 2025

DOI: 10.25728/ubs.2025.115.10