Node Removal Effect on Polarization in Social Networks
Polarization and influence in online social networks
I. Popp (TU Delft - Electrical Engineering, Mathematics and Computer Science)
A.L.D. Latour – Mentor (TU Delft - Electrical Engineering, Mathematics and Computer Science)
M. Khosla – Graduation committee member (TU Delft - Electrical Engineering, Mathematics and Computer Science)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Online social networks can facilitate ideological polarization by shaping how individuals interact with and reinforce opinions. However, the specific roles played by different types of influential nodes in accelerating or delaying polarization remain insufficiently understood.
This study investigates the effect of targeted node removal on polarization convergence time under the Dandekar opinion dynamics model. Convergence time measures the number of steps it takes for the network to reach a stable polarized state. We compare a random node removal baseline against interventions targeting boundary spanners (nodes acting as inter-community bridges) and em provincial hubs (nodes acting as intra-community leaders). We quantify these effects in both synthetic networks generated using a stochastic block model (SBM) and empirical subgraphs sampled from VK, a Russian online social networking platform. We explicitly assume a simplified setting consisting of exactly two communities with initially weakly divergent opinions of 50 nodes each, focusing on the micro-level effect.
In SBM networks, removing boundary spanners accelerated convergence by 7.69 steps relative to random removal, whereas removing provincial hubs delayed convergence by 2.73 steps. Similar patterns were observed in empirical networks, where boundary spanners removal accelerated convergence by 1.55 steps and provincial hubs removal delayed convergence by 4.93 steps.
These findings suggest that boundary spanners sustain cross-community exposure that slows polarization, whereas provincial hubs amplify within-community opinion reinforcement, promoting convergence toward ideological extremes. However, boundary spanners exert a weaker influence in empirical networks, while provincial hubs exert a stronger one. This indicates that a realistic network structure can alter node impact. Future work should investigate whether these effects persist under alternative opinion-dynamics models, more nuanced representations of social influence, and larger-scale graphs.