Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pacing Opinion Polarization via Graph Reinforcement Learning

Mingkai Liao

TL;DR

PACIFIER reformulates the canonical ModerateInternal (MI) and ModerateExpressed (ME) problems as sequential decision-making tasks, enabling adaptive intervention policies without repeated steady-state recomputation.

Abstract

Opinion polarization in online social networks poses serious risks to social cohesion and democratic processes. Recent studies formulate polarization moderation as algorithmic intervention problems under opinion dynamics models, especially the Friedkin--Johnsen (FJ) model. However, most existing methods are tailored to specific linear settings and rely on closed-form steady-state analysis, limiting scalability, flexibility, and applicability to cost-aware, nonlinear, or topology-altering interventions. We propose PACIFIER, a graph reinforcement learning framework for sequential polarization moderation via network interventions. PACIFIER reformulates the canonical ModerateInternal (MI) and ModerateExpressed (ME) problems as sequential decision-making tasks, enabling adaptive intervention policies without repeated steady-state recomputation. The framework is objective-agnostic and extends naturally to FJ-consistent settings, including budget-aware interventions, continuous internal opinions, biased-assimilation dynamics, and node removal. Extensive experiments on real-world networks demonstrate strong performance and scalability across diverse moderation scenarios.

Pacing Opinion Polarization via Graph Reinforcement Learning

TL;DR

PACIFIER reformulates the canonical ModerateInternal (MI) and ModerateExpressed (ME) problems as sequential decision-making tasks, enabling adaptive intervention policies without repeated steady-state recomputation.

Abstract

Opinion polarization in online social networks poses serious risks to social cohesion and democratic processes. Recent studies formulate polarization moderation as algorithmic intervention problems under opinion dynamics models, especially the Friedkin--Johnsen (FJ) model. However, most existing methods are tailored to specific linear settings and rely on closed-form steady-state analysis, limiting scalability, flexibility, and applicability to cost-aware, nonlinear, or topology-altering interventions. We propose PACIFIER, a graph reinforcement learning framework for sequential polarization moderation via network interventions. PACIFIER reformulates the canonical ModerateInternal (MI) and ModerateExpressed (ME) problems as sequential decision-making tasks, enabling adaptive intervention policies without repeated steady-state recomputation. The framework is objective-agnostic and extends naturally to FJ-consistent settings, including budget-aware interventions, continuous internal opinions, biased-assimilation dynamics, and node removal. Extensive experiments on real-world networks demonstrate strong performance and scalability across diverse moderation scenarios.
Paper Structure (90 sections, 2 theorems, 41 equations, 21 figures, 4 tables)

This paper contains 90 sections, 2 theorems, 41 equations, 21 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Paper Organization
  4. Related Work
  5. Opinion Polarization and Moderation
  6. Opinion polarization.
  7. The MI/ME moderation problems.
  8. Algorithms and limitations under the FJ model.
  9. Depolarization via exposure and topology modification.
  10. Traditional Network Intervention Algorithms
  11. Graph Reinforcement Learning on Networks
  12. Research Gap and Our Contribution
  13. Problem Formulation: MI/ME Moderation under the Friedkin--Johnsen Paradigm: One-Shot Planning and ANP Evaluation
  14. Friedkin-Johnsen Opinion Dynamics Model
  15. Polarization Index
  16. ...and 75 more sections

Key Result

Theorem 3.1

The ModerateInternal problem is NP-hard.

Figures (21)

  • Figure 1: Dataset-level relationships between initial polarization $\pi(\mathbf{z}^{(0)})$ and two structural indicators across the 31 topic graphs. Higher cross-camp mixing (larger $\rho_{\pm}$) and higher connectivity (larger $\bar{d}$) are associated with lower initial polarization. Pearson and Spearman statistics are reported in the plot titles.
  • Figure 2: Synthetic benchmark results across six node-size ranges (30--50, 50--100, 100--200, 200--300, 300--400, 400--500), with $100$ graphs per range. Bars compare PACIFIER with all baselines on four tasks: MI, MI-cost, ME, and ME-cost. Lower is better if the reported metric is polarization/ANP; higher is better if it is a score (%). (See the main text for the exact metric definition used in these plots.)
  • Figure 3: Real datasets (MI): per-dataset bar comparison (lower is better).
  • Figure 4: Real datasets (MI-cost): per-dataset bar comparison (lower is better).
  • Figure 5: Real datasets (ME): per-dataset bar comparison (lower is better).
  • ...and 16 more figures

Theorems & Definitions (4)

  • Definition 3.1: Polarization Index
  • Theorem 3.1: matakos2017measuring
  • Theorem 3.2: matakos2017measuring
  • Definition 4.1: Topology-preserving intervention