pith. sign in

arxiv: 2602.23390 · v3 · submitted 2026-02-23 · 💻 cs.SI · cs.LG

PACIFIER: Pacing Opinion Depolarization via a Unified Graph Learning Framework

Mingkai Liao This is my paper

Pith reviewed 2026-05-15 20:21 UTC · model grok-4.3

classification 💻 cs.SI cs.LG
keywords opinion polarizationFriedkin-Johnsen modelgraph reinforcement learningsocial network interventiondepolarizationsequential planningTwitter networksscale transfer
0
0 comments X

The pith

Opinion polarization can be moderated by reformulating it as sequential graph intervention planning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that polarization moderation under the Friedkin-Johnsen model, long handled through repeated analytical equilibrium calculations, can be recast as a graph-based sequential planning task. PACIFIER trains a unified graph learning and reinforcement learning system on synthetic graphs under 50 nodes and applies it to real networks of up to 155k nodes by using four scale-compatible designs. The system produces intervention sequences scored by accumulated normalized polarization and supports cost-aware, continuous, and topology-changing actions. If the transfer works, practical moderation becomes feasible at social media scale without recomputing steady states at every step. Readers should care because current analytical methods cannot keep pace with the size of actual online networks.

Core claim

PACIFIER reformulates minimum intervention and maximum effort problems for the Friedkin-Johnsen model as ordered graph-intervention tasks evaluated by accumulated normalized polarization. It introduces PACIFIER-RL for long-horizon value learning and PACIFIER-Greedy for myopic ranking, using a two-echo-chamber training distribution, anchor-and-mark history encoding, normalized global features, and residual-polarization rewards to achieve reliable transfer from small synthetic graphs to large real-world networks.

What carries the argument

The PACIFIER framework that converts FJ moderation into sequential graph-intervention planning scored by accumulated normalized polarization, enabled by four scale-compatible designs for small-to-large transfer.

Load-bearing premise

The four scale-compatible designs enable reliable transfer from synthetic graphs under 50 nodes to large real-world networks without major performance loss.

What would settle it

Applying the trained model to a new large Twitter network and obtaining minimum intervention scores substantially below those of analytical solvers would falsify the transfer claim.

Figures

Figures reproduced from arXiv: 2602.23390 by Mingkai Liao.

Figure 1
Figure 1. Figure 1: Illustration of FJ opinion iteration under two contrasting polarization regimes. The top panel starts from a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the graph-intervention taxonomy discussed in Sec. 2.2. Panels (a) and (b) correspond to [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Offline training workflow of PACIFIER. The Two-Echo-Chamber Instance Builder first generates synthetic [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Online application of PACIFIER on a real-world network. At each step, the encoder [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: State observability across graph-intervention regimes. In structure-disrupting intervention, node removal [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of two complementary sources of state aliasing in topology-preserving continuous-opinion [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Synthetic benchmark results across six node-size ranges (30–50, 50–100, 100–200, 200–300, 300–400, [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Real datasets (MI): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Real datasets (ME): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Real datasets (node_removal): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Real datasets: heatmap summaries for MI, ME, and node_removal. Lower is better. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Real datasets (MI): polarization trajectories on three representative datasets (small / medium / large by node [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Real datasets (ME): polarization trajectories on three representative datasets (small / medium / large by node [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Real datasets (node_removal): polarization trajectories on three representative datasets (small / medium / [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Real datasets (continuous-MI): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p023_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Real datasets (continuous-ME): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p023_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Real datasets (cost-MI): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p024_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Real datasets (cost-ME): per-dataset bar comparison (lower is better). [PITH_FULL_IMAGE:figures/full_fig_p024_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Qualitative illustration on the follow_germanwings dataset under cost-ME at intervention progress 0.1. and ranking adds O(n log n). Thus, deployment-time complexity is O(K(n + m) + n log n), which is near-linear on sparse graphs for small fixed K [PITH_FULL_IMAGE:figures/full_fig_p025_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Qualitative illustration on the follow_germanwings dataset under node_removal at intervention progress 0.1. 30-50 50-100 100-200 200-300 300-400 400-500 Range 0.14 0.16 0.18 0.20 0.22 0.24 0.26 Score (%) Synthetic (MI) Method ExtremeExpressed-FI ExtremeNeighbours-FI Greedy PACIFIER-Greedy PACIFIER-Greedy-FI PACIFIER-RL PACIFIER-RL-FI (a) MI on synthetic graphs under full￾information comparison. 30-50 50-1… view at source ↗
Figure 21
Figure 21. Figure 21: Synthetic full-information comparison across three representative settings: [PITH_FULL_IMAGE:figures/full_fig_p026_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Runtime comparison on real-world datasets under the standard setting. Left: all methods are compared [PITH_FULL_IMAGE:figures/full_fig_p027_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Runtime comparison on synthetic graphs under the full-information setting. Left: all methods are included. [PITH_FULL_IMAGE:figures/full_fig_p027_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Dataset-level relationships between initial polarization [PITH_FULL_IMAGE:figures/full_fig_p030_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Complete real-world polarization trajectories for MI over all retained datasets. Lower and earlier is better. [PITH_FULL_IMAGE:figures/full_fig_p033_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Complete real-world polarization trajectories for [PITH_FULL_IMAGE:figures/full_fig_p034_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Complete real-world polarization trajectories for [PITH_FULL_IMAGE:figures/full_fig_p035_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: Heatmap summaries for the four extended real-world settings. Each heatmap reports ANP/AUC scores [PITH_FULL_IMAGE:figures/full_fig_p037_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Real datasets under continuous-MI: complete polarization trajectories over all retained datasets. Lower and earlier is better. 39 [PITH_FULL_IMAGE:figures/full_fig_p039_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Real datasets under continuous-ME: complete polarization trajectories over all retained datasets. Lower and earlier is better. 40 [PITH_FULL_IMAGE:figures/full_fig_p040_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Real datasets under cost-MI: complete polarization trajectories over all retained datasets. Lower and earlier is better. 41 [PITH_FULL_IMAGE:figures/full_fig_p041_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Real datasets under cost-ME: complete polarization trajectories over all retained datasets. Lower and earlier is better. 42 [PITH_FULL_IMAGE:figures/full_fig_p042_32.png] view at source ↗
read the original abstract

PACIFIER: Pacing Opinion Depolarization via a Unified Graph Learning Framework Opinion polarization moderation under the Friedkin-Johnsen (FJ) model is typically treated as an analytical optimization problem. Existing algorithms rely on linear steady-state analysis and repeated equilibrium recomputation, leading to poor scalability and limited adaptability to rich intervention regimes. This paper explores whether polarization moderation can be reformulated as a graph-based sequential planning problem. We propose PACIFIER, the first unified graph-learning and graph reinforcement learning framework for FJ-based intervention. It reformulates canonical MI and ME problems as ordered graph-intervention tasks evaluated by Accumulated Normalized Polarization (ANP). The framework includes PACIFIER-RL for long-horizon value learning and PACIFIER-Greedy for efficient myopic ranking, supporting cost-aware moderation, continuous opinions, and topology-altering node removal. The core challenge is small-to-large transfer. PACIFIER is trained on synthetic graphs with fewer than 50 nodes but must generalize to large real-world networks. To achieve this, we integrate four scale-compatible designs: a two-echo-chamber training distribution, anchor-and-mark history encoding, normalized global features, and residual-polarization rewards. These components make topology-preserving FJ moderation observable and learnable across graph scales. Experiments on 15 real-world Twitter networks (up to 155,599 nodes) show that PACIFIER matches analytical solvers in MI and consistently outperforms baselines in ME, continuous-ME, cost-ME, and node removal. PACIFIER-RL proves especially effective when long-horizon costs or structural consequences dominate immediate gains.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes PACIFIER, the first unified graph-learning and graph reinforcement learning framework for moderating opinion polarization under the Friedkin-Johnsen (FJ) model. It reformulates minimum intervention (MI) and maximum entropy (ME) problems as sequential graph-intervention tasks evaluated by a new Accumulated Normalized Polarization (ANP) metric. PACIFIER-RL handles long-horizon value learning while PACIFIER-Greedy provides efficient myopic ranking; both support cost-aware moderation, continuous opinions, and node removal. The framework is trained exclusively on synthetic graphs with fewer than 50 nodes using four scale-compatible designs (two-echo-chamber training distribution, anchor-and-mark history encoding, normalized global features, residual-polarization rewards) and is evaluated on 15 real-world Twitter networks up to 155,599 nodes, claiming to match analytical solvers on MI and outperform baselines on ME variants.

Significance. If the scale-transfer results are substantiated, the work would provide a scalable learning-based alternative to repeated linear steady-state optimization for FJ interventions, enabling richer regimes (cost, continuous, structural) on networks too large for analytical methods. The unified RL-plus-greedy formulation and ANP metric constitute concrete contributions that could influence computational social science approaches to depolarization.

major comments (2)
  1. [Abstract] Abstract: the central claim that the four scale-compatible designs enable reliable transfer from <50-node synthetics to 155k-node Twitter graphs without major performance loss is load-bearing, yet no ablation that removes each design individually and re-evaluates the resulting policy on the real-world instances (especially the largest graph) is reported; without this evidence the observed parity with analytical MI solvers and gains on ME variants could arise from other factors such as graph structure or the ANP definition itself.
  2. [Experiments] Experiments section: the abstract states that PACIFIER 'matches analytical solvers in MI and consistently outperforms baselines' on 15 networks, but provides neither error bars, standard deviations, nor statistical significance tests for any of the reported metrics (MI, ME, continuous-ME, cost-ME, node removal), rendering the empirical superiority unverifiable from the given text.
minor comments (1)
  1. [Abstract] Abstract: the precise mathematical definition of Accumulated Normalized Polarization (ANP) is not supplied, which is needed to understand how the metric aggregates polarization across intervention steps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment point by point below and will incorporate revisions to strengthen the empirical support for our claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the four scale-compatible designs enable reliable transfer from <50-node synthetics to 155k-node Twitter graphs without major performance loss is load-bearing, yet no ablation that removes each design individually and re-evaluates the resulting policy on the real-world instances (especially the largest graph) is reported; without this evidence the observed parity with analytical MI solvers and gains on ME variants could arise from other factors such as graph structure or the ANP definition itself.

    Authors: We agree that individual ablations of each of the four scale-compatible designs would provide stronger evidence that they are necessary for reliable small-to-large transfer. In the revised manuscript we will add a dedicated ablation study that removes each component in turn (two-echo-chamber training distribution, anchor-and-mark history encoding, normalized global features, residual-polarization rewards) and re-evaluates the resulting policies on all 15 real-world Twitter networks, with explicit reporting for the largest graph (155,599 nodes). These additional results will isolate the contribution of each design and rule out alternative explanations such as graph structure or the ANP metric itself. revision: yes

  2. Referee: [Experiments] Experiments section: the abstract states that PACIFIER 'matches analytical solvers in MI and consistently outperforms baselines' on 15 networks, but provides neither error bars, standard deviations, nor statistical significance tests for any of the reported metrics (MI, ME, continuous-ME, cost-ME, node removal), rendering the empirical superiority unverifiable from the given text.

    Authors: We acknowledge that the lack of error bars, standard deviations, and statistical tests makes the reported superiority difficult to verify. In the revision we will recompute all metrics over multiple independent runs (different random seeds) and report means together with standard deviations for every setting (MI, ME, continuous-ME, cost-ME, node removal) across the 15 networks. We will also include appropriate statistical significance tests (paired t-tests or Wilcoxon signed-rank tests with p-values) comparing PACIFIER variants against both analytical solvers and baselines. revision: yes

Circularity Check

0 steps flagged

No circularity: PACIFIER introduces novel scale-transfer designs and evaluates them independently on real networks

full rationale

The paper reformulates FJ moderation as a graph RL planning task and introduces four explicit scale-compatible components (two-echo-chamber distribution, anchor-and-mark encoding, normalized global features, residual-polarization rewards) to enable training on <50-node synthetics and deployment on 155k-node Twitter graphs. These designs are presented as new engineering choices rather than derived from or fitted to the target performance metrics. No equations reduce a claimed prediction to a previously fitted parameter by construction, no uniqueness theorem is imported via self-citation, and the reported MI/ME parity is measured on held-out real-world instances. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Review based on abstract only; no explicit free parameters, axioms, or invented entities beyond the new ANP metric are detailed. The approach relies on standard graph RL machinery plus four custom designs whose independence from prior FJ solvers is asserted but not derived.

invented entities (1)
  • Accumulated Normalized Polarization (ANP) no independent evidence
    purpose: Metric for evaluating ordered graph-intervention sequences
    Introduced as the evaluation criterion for the reformulated tasks; no external validation or derivation from prior polarization measures is provided in the abstract.

pith-pipeline@v0.9.0 · 5580 in / 1172 out tokens · 33246 ms · 2026-05-15T20:21:07.584219+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 1 internal anchor

  1. [1]

    Sunstein

    Cass R. Sunstein. The law of group polarization.The Journal of Political Philosophy, 10(2):175–195, 2002

    work page 2002

  2. [2]

    Kelly Garrett

    R. Kelly Garrett. Echo chambers online?: Politically motivated selective exposure among internet news users. Journal of Computer-Mediated Communication, 14(2):265–285, 2009

    work page 2009

  3. [3]

    Eytan Bakshy, Solomon Messing, and Lada A. Adamic. Exposure to ideologically diverse news and opinion on facebook.Science, 348(6239):1130–1132, 2015

    work page 2015

  4. [4]

    Conover, Jacob Ratkiewicz, Matthew Francisco, Bruno Gonçalves, Alessandro Flammini, and Filippo Menczer

    Michael D. Conover, Jacob Ratkiewicz, Matthew Francisco, Bruno Gonçalves, Alessandro Flammini, and Filippo Menczer. Political polarization on twitter. InProceedings of the Fifth International AAAI Conference on Weblogs and Social Media, ICWSM ’11, 2011

    work page 2011

  5. [5]

    Adamic and Natalie Glance

    Lada A. Adamic and Natalie Glance. The political blogosphere and the 2004 u.s. election: Divided they blog. In Proceedings of the 3rd International Workshop on Link Discovery, LinkKDD ’05, pages 36–43. ACM, 2005

    work page 2004

  6. [6]

    Quantifying controversy on social media.ACM Transactions on Social Computing, 1(1):3:1–3:27, 2018

    Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, and Michael Mathioudakis. Quantifying controversy on social media.ACM Transactions on Social Computing, 1(1):3:1–3:27, 2018

    work page 2018

  7. [7]

    Social influence and opinions.Journal of Mathematical Sociology, 15(3-4):193–206, 1990

    Noah E Friedkin and Eugene C Johnsen. Social influence and opinions.Journal of Mathematical Sociology, 15(3-4):193–206, 1990

    work page 1990

  8. [8]

    Morris H. DeGroot. Reaching a consensus.Journal of the American Statistical Association, 69(345):118–121, 1974

    work page 1974

  9. [9]

    How bad is forming your own opinion?Games and Economic Behavior, 92:248–265, 2015

    David Bindel, Jon Kleinberg, and Sigal Oren. How bad is forming your own opinion?Games and Economic Behavior, 92:248–265, 2015

    work page 2015

  10. [10]

    Measuring and moderating opinion polarization in social networks.Data Mining and Knowledge Discovery, 31(5):1480–1505, 2017

    Antonis Matakos, Evimaria Terzi, and Panayiotis Tsaparas. Measuring and moderating opinion polarization in social networks.Data Mining and Knowledge Discovery, 31(5):1480–1505, 2017

    work page 2017

  11. [11]

    Tsourakakis

    Cameron Musco, Christopher Musco, and Charalampos E. Tsourakakis. Minimizing polarization and disagreement in social networks. InProceedings of the 2018 World Wide Web Conference (WWW ’18), 2018

    work page 2018

  12. [12]

    Quantifying and minimizing risk of conflict in social networks

    Xi Chen, Jefrey Lijffijt, and Tijl De Bie. Quantifying and minimizing risk of conflict in social networks. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 1197–1206, 2018

    work page 2018

  13. [13]

    Reducing controversy by connecting opposing views

    Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, and Michael Mathioudakis. Reducing controversy by connecting opposing views. InProceedings of the Tenth ACM International Conference on Web Search and Data Mining (WSDM ’17), 2017

    work page 2017

  14. [14]

    Minimizing polarization and disagreement in social networks via link recommendation

    Liwang Zhu, Qi Bao, and Zhongzhi Zhang. Minimizing polarization and disagreement in social networks via link recommendation. InAdvances in Neural Information Processing Systems, volume 34, pages 29047–29058, 2021

    work page 2021

  15. [15]

    A nearly-linear time algorithm for minimizing risk of conflict in social networks

    Liwang Zhu and Zhongzhi Zhang. A nearly-linear time algorithm for minimizing risk of conflict in social networks. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 2648–2656, 2022

    work page 2022

  16. [16]

    Rácz and Daniel E

    Miklos Z. Rácz and Daniel E. Rigobon. Towards consensus: Reducing polarization by perturbing social networks. IEEE Transactions on Network Science and Engineering, 2023

    work page 2023

  17. [17]

    Minimizing polarization in noisy leader-follower opinion dynamics

    Wanyue Xu and Zhongzhi Zhang. Minimizing polarization in noisy leader-follower opinion dynamics. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pages 2856–2865, 2023

    work page 2023

  18. [18]

    Online minimization of polarization and disagreement via low-rank matrix bandits

    Federico Cinus, Yuko Kuroki, Atsushi Miyauchi, and Francesco Bonchi. Online minimization of polarization and disagreement via low-rank matrix bandits. InInternational Conference on Learning Representations, 2026. 43 Running Title for Header Table 6: Full ablation results under MI with PACIFIER-Greedy on real-world datasets. Method follow_germanwings follo...

    work page 2026

  19. [19]

    Rusu, Joel Veness, Marc G

    V olodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis Hassabis. Human- level control through deep reinforcement...

    work page 2015

  20. [20]

    Khalil, Yuyu Zhang, Bistra Dilkina, and Le Song

    Hanjun Dai, Elias B. Khalil, Yuyu Zhang, Bistra Dilkina, and Le Song. Learning combinatorial optimization algorithms over graphs. InAdvances in Neural Information Processing Systems, volume 30, 2017

    work page 2017

  21. [21]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations, 2017

    work page 2017

  22. [22]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InAdvances in Neural Information Processing Systems, volume 30, pages 1024–1034, 2017

    work page 2017

  23. [23]

    Finding key players in complex networks through deep reinforcement learning.Nature Machine Intelligence, 2(6):317–324, 2020

    Changjun Fan, Li Zeng, Yizhou Sun, and Yang-Yu Liu. Finding key players in complex networks through deep reinforcement learning.Nature Machine Intelligence, 2(6):317–324, 2020

    work page 2020

  24. [24]

    Gcomb: Learning budget-constrained combinatorial algorithms over billion-sized graphs

    Sahil Manchanda, Akash Mittal, Anuj Dhawan, Sourav Medya, Sayan Ranu, and Ambuj Singh. Gcomb: Learning budget-constrained combinatorial algorithms over billion-sized graphs. InAdvances in Neural Information Processing Systems, volume 33, pages 20000–20011, 2020

    work page 2020

  25. [25]

    Maximizing the spread of influence through a social network

    David Kempe, Jon Kleinberg, and Éva Tardos. Maximizing the spread of influence through a social network. In Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, pages 137–146, 2003

    work page 2003

  26. [26]

    Thai, Renhao Xue, James Song, Meikang Qiu, and Liang Zhao

    Chen Ling, Junji Jiang, Junxiang Wang, My T. Thai, Renhao Xue, James Song, Meikang Qiu, and Liang Zhao. Deep graph representation learning and optimization for influence maximization. InProceedings of the 40th International Conference on Machine Learning, 2023

    work page 2023

  27. [27]

    Opinion de-polarization in social networks with GNNs

    Nicholas Mylonas and Thanassis Spyropoulos. Opinion depolarization in social networks with graph neural networks.arXiv preprint arXiv:2412.09404, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  28. [28]

    Opinion optimization in directed social networks

    Haoxin Sun and Zhongzhi Zhang. Opinion optimization in directed social networks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 4623–4630, 2023

    work page 2023

  29. [29]

    Flaviano Morone and Hernán A. Makse. Influence maximization in complex networks through optimal percolation. Nature, 524(7563):65–68, 2015

    work page 2015

  30. [30]

    Network dismantling.Proceed- ings of the National Academy of Sciences, 113(44):12368–12373, 2016

    Alfredo Braunstein, Luca Dall’Asta, Guilhem Semerjian, and Lenka Zdeborová. Network dismantling.Proceed- ings of the National Academy of Sciences, 113(44):12368–12373, 2016

    work page 2016

  31. [31]

    Growing well-connected graphs

    Arpita Ghosh and Stephen Boyd. Growing well-connected graphs. InProceedings of the 45th IEEE Conference on Decision and Control, pages 6605–6611, 2006

    work page 2006

  32. [32]

    Opinion dynamics and bounded confidence models, analysis, and simulation.Journal of Artificial Societies and Social Simulation, 5(3):2, 2002

    Rainer Hegselmann and Ulrich Krause. Opinion dynamics and bounded confidence models, analysis, and simulation.Journal of Artificial Societies and Social Simulation, 5(3):2, 2002

    work page 2002

  33. [33]

    Mixing beliefs among interacting agents.Advances in Complex Systems, 3(1–4):87–98, 2000

    Guillaume Deffuant, David Neau, Frederic Amblard, and Gerard Weisbuch. Mixing beliefs among interacting agents.Advances in Complex Systems, 3(1–4):87–98, 2000

    work page 2000

  34. [34]

    Pranav Dandekar, Ashish Goel, and David T. Lee. Biased assimilation, homophily, and the dynamics of polarization.Proceedings of the National Academy of Sciences, 110(15):5791–5796, 2013

    work page 2013

  35. [35]

    Consensus problems on networks with antagonistic interactions.IEEE Transactions on Automatic Control, 58(4):935–946, 2013

    Claudio Altafini. Consensus problems on networks with antagonistic interactions.IEEE Transactions on Automatic Control, 58(4):935–946, 2013. 45

    work page 2013