arxiv: 2604.06600 · v2 · submitted 2026-04-08 · 💻 cs.SI

Recognition: unknown

IntervenSim: Intervention-Aware Social Network Simulation for Opinion Dynamics

Yunyao Zhang , Zuocheng Ying , Xinglang Zhang , Junqing Yu , Peng Fang , Xu Chen , Wei Yang , Zikai Song

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:29 UTC · model grok-4.3

classification 💻 cs.SI

keywords social network simulationopinion dynamicsintervention modelingLLM agentsevent evolutionclosed-loop simulationmulti-agent systemsfeedback mechanisms

0 comments

The pith

IntervenSim models social events by coupling source interventions with crowd reactions in a continuous feedback loop.

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

Prior simulations of social networks assume an event follows a fixed path once set in motion, but real online events involve ongoing source interventions and collective interactions that reshape trajectories and can trigger secondary waves of attention or attitude changes. IntervenSim addresses this by using source agents to generate and adjust interventions while crowd agents model group reactions, linking the two so each round of simulation feeds back into the next. The result is a closed-loop process that lets the model adapt simulated behaviors based on prior outcomes rather than running a single static trajectory. A sympathetic reader would care because this setup produces closer matches to observed popularity curves and opinion shifts on real events.

Core claim

The paper introduces an intervention-aware simulation framework, IntervenSim, that models event evolution and intervention in a closed loop. Source agents handle event developments and source-side interventions while crowd agents handle collective reactions, with their co-evolution captured through an intervention-aware mechanism that couples source-side intervention, group interaction, and feedback-driven adjustment of subsequent interventions.

What carries the argument

The intervention-aware mechanism that couples source-side intervention, group interaction, and feedback-driven adjustment of subsequent interventions.

If this is right

Simulates regular event trajectories more faithfully than static models.
Captures opinion dynamics under intervention in complex cases with secondary explosions or shifts.
Achieves 41.6 percent lower MAPE and 66.9 percent lower DTW error on diverse real events.
Reduces computational cost by using fewer but more capable agents.
Enables closed-loop adjustment where each intervention round responds to simulated crowd feedback.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could support pre-testing of proposed interventions by running them inside the simulator before real deployment.
It opens the possibility of scaling the same loop structure to study how interventions in one topic spill over to related events through shared crowd agents.
Replacing large numbers of simple agents with a smaller set of LLM agents may generalize to other domains where continuous feedback between actors and environment matters.
Controlled experiments with human participants could directly measure how closely the agent outputs match actual intervention responses.

Load-bearing premise

LLM-powered source and crowd agents can faithfully reproduce real human intervention behaviors and collective opinion dynamics without introducing systematic biases or artifacts from the models themselves.

What would settle it

Run IntervenSim and prior static simulators on a fresh real-world event with documented interventions, then compare both sets of predicted popularity and opinion time series against the actual observed data to check whether the reported error reductions appear.

Figures

Figures reproduced from arXiv: 2604.06600 by Junqing Yu, Peng Fang, Wei Yang, Xinglang Zhang, Xu Chen, Yunyao Zhang, Zikai Song, Zuocheng Ying.

**Figure 1.** Figure 1: Comparison of social simulation paradigms. Traditional ABMs (left) predict outcomes but lack behavioral heterogeneity [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: IntervenSim overview. The framework comprises two interactive processes that capture top-down intervention and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Dynamic group generation process. The system first [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Time cost comparison between GA-S3 and IntervenSim across different stages. Circles represent GA-S3 , stars represent IntervenSim, solid lines indicate per-agent time, and dashed lines indicate per-event time. evaluating the model’s ability to capture realistic fluctuation and trend dynamics. • Z-score (Reproducibility) [10]: Reports the stability of results across repeated simulations. We run each config… view at source ↗

**Figure 5.** Figure 5: Heatmap of five intervention and comment types [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Traffic trend comparison under different ablation settings across events. Results show that interventions, their timing, [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 8.** Figure 8: Overall intuitive impact of our framework. The results demonstrate the advantages of IntervenSim, including high [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 10.** Figure 10: Event ID Domain Intervention Description Distinctive Features #2 Education Publicity & Announcement A school cafeteria was reported to have served spoiled pork. A traffic surge and two explosive events of source agent intervention #7 Education Response A case of academic dishonesty led to a professor’s dismissal. A high-profile event with a viewership peak on the 3rd day. #14 Education Response 1,477 fres… view at source ↗

**Figure 9.** Figure 9: Additional event-level traffic trends under different intervention and control settings, validating the consistency of [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

read the original abstract

LLM-based social network simulation introduces a new computational approach for modeling event evolution in complex online environments. However, existing methods typically simulate social processes under a fixed event trajectory, treating the event as static once initialized and overlooking intervention dynamics, and thus fail to capture the intrinsic evolution of real social network events, where source-side interventions and collective interactions continuously reshape event trajectories, sometimes leading to secondary popularity explosions and collective attitude shifts. To address this limitation, we introduce an intervention-aware simulation framework, IntervenSim, that models event evolution and intervention in a closed loop. We model event developments and source-side interventions using source agents, and collective crowd reactions using crowd agents, capturing their continuous co-evolution through an intervention-aware mechanism that couples source-side intervention, group interaction, and feedback-driven adjustment of subsequent interventions. Experiments on diverse real-world events show that IntervenSim improves MAPE by 41.6% and DTW by 66.9% over prior frameworks, while reducing computational cost with fewer yet more capable agents. These improvements indicate that IntervenSim not only simulates regular event trajectories more faithfully, but also better captures opinion dynamics under intervention in complex cases.

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.

Desk Editor's Note private letter to a colleague

IntervenSim adds a closed-loop intervention layer to LLM social simulations, but the accuracy gains rest on untested assumptions about how well the agents match real human behavior.

read the letter

The paper's main move is to replace fixed event trajectories with a loop where source agents produce interventions and crowd agents generate collective reactions, with feedback that lets each side reshape the next round. This coupling is the concrete difference from the static simulation baselines they reference, and it targets real patterns like secondary spikes or attitude shifts in online events. That framing is a direct response to a limitation in prior work, and the architecture is straightforward to follow once the agent roles are laid out. The reported results claim solid trajectory matches on diverse real-world cases along with lower compute from using fewer agents. If the full experiments include proper baselines, multiple datasets, and controls, the MAPE and DTW numbers would be useful evidence for the approach. The main weakness is the unexamined premise that current LLMs can stand in for human interveners and responders without introducing their own biases or artifacts. The abstract supplies no ablations across models, no human calibration data, and no checks on prompt sensitivity or long-horizon coherence, so the gains could partly reflect LLM quirks rather than the closed-loop design itself. Experimental details on how intervention signals are extracted from real traces are also thin. This is aimed at researchers who build or apply agent-based models for opinion dynamics and online influence. Anyone working on simulation tools for misinformation or collective behavior would find the agent split and feedback mechanism worth examining, even if they later add their own validation steps. I would send it for peer review. The problem is relevant, the proposed mechanism is clear, and referees can focus on tightening the empirical validation.

Referee Report

2 major / 2 minor

Summary. The paper introduces IntervenSim, an intervention-aware social network simulation framework using LLM-based source agents to model event developments and source-side interventions and crowd agents to model collective reactions. It proposes a closed-loop mechanism that couples interventions, group interactions, and feedback-driven adjustments to capture the continuous co-evolution of events and opinion dynamics, in contrast to prior static-trajectory simulations. Experiments on diverse real-world events are reported to yield 41.6% MAPE and 66.9% DTW improvements over prior frameworks while using fewer agents.

Significance. If the empirical gains are reproducible and the LLM agents prove faithful to human behavior, the work would meaningfully advance computational modeling of dynamic social processes by addressing intervention effects that existing methods overlook. The closed-loop design and agent-based architecture represent a concrete step toward more realistic simulation of secondary popularity spikes and attitude shifts.

major comments (2)

[Experiments / Results] The headline empirical claim (41.6% MAPE / 66.9% DTW gains) is load-bearing for the paper's contribution, yet the abstract and experimental description supply no information on the specific prior frameworks used as baselines, the real-world event datasets, how intervention signals were extracted, statistical significance tests, error bars, or ablation controls. Without these details the reported improvements cannot be evaluated for robustness.
[Methodology / Agent Design] The central modeling assumption—that LLM-powered source and crowd agents faithfully reproduce human intervention behaviors and collective opinion dynamics without systematic artifacts—is untested in the manuscript. No ablations across LLMs, no human-subject calibration, and no analysis of prompt sensitivity or long-horizon coherence are provided, leaving open the possibility that the observed gains arise from model idiosyncrasies rather than the closed-loop architecture.

minor comments (2)

[Abstract] The abstract refers to 'diverse real-world events' without naming them or describing the data sources; adding a brief table or list would improve clarity.
[Framework Description] Notation for the intervention-aware coupling mechanism (source intervention ↔ crowd feedback) should be formalized with a diagram or pseudocode to make the closed-loop structure explicit.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below, providing honest clarifications and committing to revisions that strengthen the work without misrepresenting our contributions.

read point-by-point responses

Referee: [Experiments / Results] The headline empirical claim (41.6% MAPE / 66.9% DTW gains) is load-bearing for the paper's contribution, yet the abstract and experimental description supply no information on the specific prior frameworks used as baselines, the real-world event datasets, how intervention signals were extracted, statistical significance tests, error bars, or ablation controls. Without these details the reported improvements cannot be evaluated for robustness.

Authors: We agree that the experimental reporting in the initial submission was insufficiently detailed to allow full evaluation of the claims. In the revised manuscript, we will expand the Experiments section to explicitly name and describe the prior static-trajectory frameworks used as baselines, provide full details on the real-world event datasets (including sources, sizes, and selection criteria), explain the extraction of intervention signals from the data, report statistical significance tests with p-values, include error bars from multiple simulation runs, and present ablation controls isolating the closed-loop components. These additions will directly address the robustness concerns. revision: yes
Referee: [Methodology / Agent Design] The central modeling assumption—that LLM-powered source and crowd agents faithfully reproduce human intervention behaviors and collective opinion dynamics without systematic artifacts—is untested in the manuscript. No ablations across LLMs, no human-subject calibration, and no analysis of prompt sensitivity or long-horizon coherence are provided, leaving open the possibility that the observed gains arise from model idiosyncrasies rather than the closed-loop architecture.

Authors: We acknowledge that the fidelity of LLM agents to human behavior is an important and untested assumption in the current version. We will revise the Methodology and Experiments sections to include ablations across multiple LLMs, sensitivity analysis for key prompts, and metrics for long-horizon coherence to demonstrate that performance gains arise from the intervention-aware closed-loop mechanism. Full human-subject calibration studies, however, fall outside the scope of this work and would require separate resources; we will explicitly discuss this as a limitation and outline it as future work while emphasizing that the empirical results on real-world events provide supporting evidence for the framework's value. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external comparisons

full rationale

The paper introduces IntervenSim as a closed-loop agent-based framework with source and crowd agents coupled by an intervention-aware mechanism. Its load-bearing claims are strictly empirical: measured MAPE and DTW improvements (41.6 % and 66.9 %) relative to prior frameworks on real-world event traces. No equations, parameter-fitting steps, self-definitions, or uniqueness theorems appear in the provided text that would reduce these performance numbers to the model's own inputs or to prior self-citations. The evaluation is presented as direct, out-of-sample comparison against external baselines, rendering the derivation chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The framework depends on the unverified assumption that LLM agents can stand in for human interveners and crowds; no free parameters or invented physical entities are mentioned, but the agent types themselves are new modeling constructs introduced without external validation.

axioms (1)

domain assumption LLM agents can accurately simulate real human intervention decisions and collective opinion shifts
Invoked to justify the closed-loop simulation of event evolution.

invented entities (2)

Source agents no independent evidence
purpose: Model event developments and source-side interventions
New component introduced to handle dynamic interventions.
Crowd agents no independent evidence
purpose: Model collective crowd reactions and interactions
New component introduced to handle group-level responses.

pith-pipeline@v0.9.0 · 5523 in / 1348 out tokens · 100970 ms · 2026-05-10T17:29:46.125868+00:00 · methodology

discussion (0)

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

OmniTrend: Content-Context Modeling for Scalable Social Popularity Prediction
cs.CV 2026-04 unverdicted novelty 6.0

OmniTrend predicts popularity by combining separate content attractiveness and contextual exposure predictors using cross-modal and exogenous signals.
HotComment: A Benchmark for Evaluating Popularity of Online Comments
cs.AI 2026-04 unverdicted novelty 6.0

HotComment is a new multimodal benchmark that quantifies online comment popularity via content quality assessment, interaction-based prediction, and agent-simulated user engagement, accompanied by the StyleCmt stylist...
Seeing Further and Wider: Joint Spatio-Temporal Enlargement for Micro-Video Popularity Prediction
cs.MM 2026-04 unverdicted novelty 5.0

A new joint spatio-temporal enlargement model for micro-video popularity prediction using frame scoring for long sequences and a topology-aware memory bank for unbounded historical associations.
CurEvo: Curriculum-Guided Self-Evolution for Video Understanding
cs.CV 2026-04 unverdicted novelty 4.0

CurEvo integrates curriculum guidance into self-evolution to structure autonomous improvement of video understanding models, yielding gains on VideoQA benchmarks.

Reference graph

Works this paper leans on

108 extracted references · 40 canonical work pages · cited by 4 Pith papers · 12 internal anchors

[1]

Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. GPT-4 Technical Report.arXiv preprint arXiv:2303.08774 (2023)

work page internal anchor Pith review Pith/arXiv arXiv 2023
[2]

Badr AlKhamissi, Muhammad ElNokrashy, Mai Alkhamissi, and Mona Diab. 2024. Investigating Cultural Alignment of Large Language Models. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Lun-Wei Ku, Andre Martins, and Vivek Srikumar (Eds.). Association for Computational Linguistics, Bangkok, Th...

2024
[3]

Kozlowski, Bernard Koch, Erik Brynjolfsson, James Evans, and Michael S

Jacy Reese Anthis, Ryan Liu, Sean M Richardson, Austin C. Kozlowski, Bernard Koch, Erik Brynjolfsson, James Evans, and Michael S. Bernstein. 2025. Po- sition: LLM Social Simulations Are a Promising Research Method. InForty- second International Conference on Machine Learning Position Paper Track. https: //openreview.net/forum?id=cRBg1dtj7o

2025
[4]

Eytan Bakshy, Solomon Messing, and Lada A Adamic. 2015. Exposure to ideologi- cally diverse news and opinion on Facebook.Science348, 6239 (2015), 1130–1132

2015
[5]

Youssef Bouanan, Gregory Zacharewicz, Judicael Ribault, and Bruno Vallespir
[6]

Discrete Event System Specification-Based Framework for Modeling and Simulation of Propagation Phenomena in Social Networks: Application to the Information Spreading in a Multi-Layer Social Network.Simulation95, 5 (2019), 411–427

2019
[7]

Elizabeth Bruch and Jon Atwell. 2015. Agent-based models in empirical social research.Sociological methods & research44, 2 (2015), 186–221

2015
[8]

Serina Chang, Alicja Chaszczewicz, Emma Wang, Maya Josifovska, Emma Pierson, and Jure Leskovec. 2024. LLMs generate structurally realistic social networks but overestimate political homophily.arXiv preprint arXiv:2408.16629(2024)

work page arXiv 2024
[9]

Myra Cheng, Esin Durmus, and Dan Jurafsky. 2023. Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models. InPro- ceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Anna Rogers, Jordan Boyd-Graber, and Naoaki Okazaki (Eds.). Association for Computational Linguistic...

work page doi:10.18653/v1/2023.acl-long.84 2023
[10]

Bastien Chopard and Michel Droz. 1998. Cellular Automata.Modelling of Physical 1 (1998)

1998
[11]

Alexander E Curtis, Tanya A Smith, Bulat A Ziganshin, and John A Elefteriades
[12]

The mystery of the Z-score.Aorta4, 04 (2016), 124–130

2016
[13]

Chongyuan Dai, Jinpeng Hu, Hongchang Shi, Zhuo Li, Xun Yang, and Meng Wang. 2025. Psyche-R1: Towards Reliable Psychological LLMs through Unified Empathy, Expertise, and Reasoning. arXiv:2508.10848 [cs.CL] https://arxiv.org/ abs/2508.10848

work page arXiv 2025
[14]

Chongyuan Dai, Yaling Shen, Jinpeng Hu, Zihan Gao, Jia Li, Yishun Jiang, Yaxiong Wang, Liu Liu, and Zongyuan Ge. 2026. Tears or Cheers? Benchmarking LLMs via Culturally Elicited Distinct Affective Responses. arXiv:2601.13024 [cs.CL] https://arxiv.org/abs/2601.13024

work page arXiv 2026
[15]

Xiongwei Dang, Wenxuan Liu, Xian Zhong, and Zheng Wang. 2025. SegTraj: A Segmented-Trajectory-Aware Spatio-Temporal Graph Convolutional Network for Social Group Detection. InProceedings of the 33rd ACM International Conference on Multimedia(Dublin, Ireland)(MM ’25). Association for Computing Machinery, New York, NY, USA, 4571–4579. doi:10.1145/3746027.3755506

work page doi:10.1145/3746027.3755506 2025
[16]

Paul Davidsson. 2002. Agent-Based Social Simulation: A Computer Science View. Journal of Artificial Societies and Social Simulation5, 1 (2002)

2002
[17]

2012.A Systems-Based Approach to Policymaking

Kenyon B De Greene. 2012.A Systems-Based Approach to Policymaking. Springer Science & Business Media

2012
[18]

Morris H DeGroot. 1974. Reaching a consensus.Journal of the American Statistical association69, 345 (1974), 118–121

1974
[19]

Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models.arXiv preprint arXiv:2407.21783(2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[20]

1996.Growing Artificial Societies: Social Science from the Bottom Up

Joshua M Epstein and Robert Axtell. 1996.Growing Artificial Societies: Social Science from the Bottom Up. Brookings Institution Press

1996
[21]

Anjalie Field, Doron Kliger, Shuly Wintner, Jennifer Pan, Dan Jurafsky, and Yulia Tsvetkov. 2018. Framing and Agenda-setting in Russian News: a Computational Analysis of Intricate Political Strategies. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Ellen Riloff, David Chiang, Julia Hockenmaier, and Jun’ichi Tsuji...

work page doi:10.18653/v1/d18-1393 2018
[22]

Andreas Flache, Michael Mäs, Thomas Feliciani, Edmund Chattoe-Brown, Guil- laume Deffuant, Sylvie Huet, and Jan Lorenz. 2017. Models of social influence: Towards the next frontiers.Journal of Artificial Societies and Social Simulation20, 4 (2017)

2017
[23]

Linton Freeman. 2004. The development of social network analysis.A Study in the Sociology of Science1, 687 (2004), 159–167

2004
[24]

Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2024. Large language models empowered agent-based modeling and simulation: A survey and perspectives.Humanities and Social Sciences Communications11, 1 (2024), 1–24

2024
[25]

Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, and Yong Li. 2023. S ˆ3: Social-network Simulation System with Large Language Model-Empowered Agents.arXiv preprint arXiv:2307.14984 (2023)

work page arXiv 2023
[26]

Josh A Goldstein, Jason Chao, Shelby Grossman, Alex Stamos, and Michael Tomz
[27]

How persuasive is AI-generated propaganda?PNAS nexus3, 2 (2024), pgae034

2024
[28]

Davide Grossi, Frank Dignum, Mehdi Dastani, and Lambèr Royakkers. 2005. Foundations of organizational structures in multiagent systems. InProceedings of the Fourth International Joint Conference on Autonomous Agents and Multiagent Systems(The Netherlands)(AAMAS ’05). Association for Computing Machinery, New York, NY, USA, 690–697. doi:10.1145/1082473.1082578

work page doi:10.1145/1082473.1082578 2005
[29]

Andrew M Guess, Neil Malhotra, Jennifer Pan, Pablo Barberá, Hunt Allcott, Taylor Brown, Adriana Crespo-Tenorio, Drew Dimmery, Deen Freelon, Matthew Gentzkow, et al. 2023. How do social media feed algorithms affect attitudes and behavior in an election campaign?Science381, 6656 (2023), 398–404

2023
[30]

Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, et al. 2025. Deepseek-r1 incen- tivizes reasoning in llms through reinforcement learning.Nature645, 8081 (2025), 633–638

2025
[31]

Lei Guo. 2013. Toward the third level of agenda-setting theory: A network agenda setting model 1 2. InAgenda setting in a 2.0 world. Routledge, 112–133

2013
[32]

Rainer Hegselmann. 2015. Opinion dynamics and bounded confidence: models, analysis and simulation.The Journal of Artificial Societies and Social Simulation Conference, April 2026, Preprint Yunyao Zhang et al. (2015)

2015
[33]

Yangliu Hu, Zikai Song, Na Feng, Yawei Luo, Junqing Yu, Yi-Ping Phoebe Chen, and Wei Yang. 2025. Sf2t: Self-supervised fragment finetuning of video-llms for fine-grained understanding. InProceedings of the Computer Vision and Pattern Recognition Conference. 29108–29117

2025
[34]

Hongru Ji, Yuyin Fan, Meng Zhao, Xianghua Li, Lianwei Wu, and Chao Gao
[35]

STRIDE-ED: A Strategy-Grounded Stepwise Reasoning Framework for Empathetic Dialogue Systems

STRIDE-ED: A Strategy-Grounded Stepwise Reasoning Framework for Empathetic Dialogue Systems. arXiv:2604.07100 [cs.CL] https://arxiv.org/abs/ 2604.07100

work page internal anchor Pith review Pith/arXiv arXiv
[36]

Hongru Ji, Xianghua Li, Mingxin Li, Meng Zhao, and Chao Gao. 2025. Hybrid relational graphs with sentiment-laden semantic alignment for multimodal emo- tion recognition in conversation. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence. 2973–2981

2025
[37]

2013.A solution to the ecological inference problem: Reconstructing individual behavior from aggregate data

Gary King. 2013.A solution to the ecological inference problem: Reconstructing individual behavior from aggregate data. Princeton University Press

2013
[38]

Qingchao Kong, Wenji Mao, Guandan Chen, and Daniel Zeng. 2018. Explor- ing trends and patterns of popularity stage evolution in social media.IEEE Transactions on Systems, Man, and Cybernetics: Systems50, 10 (2018), 3817–3827

2018
[39]

Liao, Esin Durmus, Alex Tamkin, and Deep Ganguli

Messi H.J. Lee, Jacob M. Montgomery, and Calvin K. Lai. 2024. Large Language Models Portray Socially Subordinate Groups as More Homogeneous, Consistent with a Bias Observed in Humans. InProceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency(Rio de Janeiro, Brazil)(FAccT ’24). Association for Computing Machinery, New York, NY,...

work page doi:10.1145/3630106.3658975 2024
[40]

Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems33 (2020), 9459–9474

2020
[41]

Mei Li, Xiang Wang, Kai Gao, and Shanshan Zhang. 2017. A survey on information diffusion in online social networks: Models and methods.Information8, 4 (2017), 118

2017
[42]

Wenbing Li, Zikai Song, Hang Zhou, Yunyao Zhang, Junqing Yu, and Wei Yang
[43]

LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing.arXiv preprint arXiv:2507.00029(2025)

work page internal anchor Pith review arXiv 2025
[44]

Wenbing Li, Hang Zhou, Junqing Yu, Zikai Song, and Wei Yang. 2024. Coupled mamba: Enhanced multimodal fusion with coupled state space model.Advances in Neural Information Processing Systems37 (2024), 59808–59832

2024
[45]

Junjie Liao, Huacong Tang, Zhou Ziheng, Yizhou Wang, and Fangwei Zhong
[46]

arXiv:2603.13876 [cs.MA] https://arxiv.org/ abs/2603.13876

How do Role Models Shape Collective Morality? Exemplar-Driven Moral Learning in Multi-Agent Simulation. arXiv:2603.13876 [cs.MA] https://arxiv.org/ abs/2603.13876

work page arXiv
[47]

2017.Public opinion

Walter Lippmann. 2017.Public opinion. Routledge

2017
[48]

Benjamin Litterer, David Jurgens, and Dallas Card. 2023. When it Rains, it Pours: Modeling Media Storms and the News Ecosystem. InFindings of the Association for Computational Linguistics: EMNLP 2023, Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational Linguistics, Singapore, 6346–6361. doi:10.18653/v1/2023.findings-emnlp.420

work page doi:10.18653/v1/2023.findings-emnlp.420 2023
[49]

Yijun Liu, Wu Liu, Xiaoyan Gu, Allen He, Weiping Wang, and Yongdong Zhang
[50]

arXiv preprint arXiv:2512.02533(2025)

PopSim: Social Network Simulation for Social Media Popularity Prediction. arXiv preprint arXiv:2512.02533(2025)

work page arXiv 2025
[51]

Yunbo Long, Yunhan Liu, and Liming Xu. 2026. EmoMAS: Emotion-Aware Multi-Agent System for High-Stakes Edge-Deployable Negotiation with Bayesian Orchestration. arXiv:2604.07003 [cs.AI] https://arxiv.org/abs/2604.07003

work page internal anchor Pith review Pith/arXiv arXiv 2026
[52]

Peng Lu and Shizhao Nie. 2019. The strength distribution and combined duration prediction of online collective actions: Big data analysis and BP neural networks. Physica A: Statistical Mechanics and its Applications535 (2019), 121031

2019
[53]

Peng Lu, Shizhao Nie, Zheng Wang, Ziwei Jing, Jianwu Yang, Zhongxiang Qi, and Wangmo Pujia. 2018. Predicting durations of online collective actions based on Peaks’ heights.Communications in Nonlinear Science and Numerical Simulation 55 (2018), 338–354

2018
[54]

Peng Lu, Zheng Wang, Shizhao Nie, Wangmo Pujia, Pengfei Lu, and Baosheng Chen. 2018. Exploring the participate propensity in cyberspace collective actions: The 5‰rule.Physica A: Statistical Mechanics and Its Applications503 (2018), 582–590

2018
[55]

Peng Lu, Hou Yang, Hao Li, Mengdi Li, and Zhuo Zhang. 2021. Swarm intelli- gence, social force and multi-agent modeling of heroic altruism behaviors under collective risks.Knowledge-Based Systems214 (2021), 106725

2021
[56]

Maxwell E McCombs and Donald L Shaw. 1972. The agenda-setting function of mass media.Public opinion quarterly36, 2 (1972), 176–187

1972
[57]

Maxwell E McCombs and Donald L Shaw. 1993. The evolution of agenda-setting research: Twenty-five years in the marketplace of ideas.Journal of communication 43, 2 (1993), 58–67

1993
[58]

1974.Dynamics of Growth in a Finite World

Dennis L Meadows, William W Behrens, Donella H Meadows, Roger F Naill, Jørgen Randers, and Erich Zahn. 1974.Dynamics of Growth in a Finite World. Vol. 360. Wright-Allen Press, Cambridge, MA

1974
[59]

Qirui Mi, Mengyue Yang, Xiangning Yu, Zhiyu Zhao, Cheng Deng, Bo An, Haifeng Zhang, Xu Chen, and Jun Wang. 2025. MF-LLM: Simulating Population Decision Dynamics via a Mean-Field Large Language Model Framework.arXiv preprint arXiv:2504.21582(2025)

work page arXiv 2025
[60]

Meinard Müller. 2007. Dynamic time warping.Information retrieval for music and motion(2007), 69–84

2007
[61]

Mark EJ Newman. 2002. Spread of epidemic disease on networks.Physical review E66, 1 (2002), 016128

2002
[62]

Shizhao Nie, Zheng Wang, Wangmo Pujia, Yuan Nie, and Peng Lu. 2018. Big data prediction of durations for online collective actions based on peak’s timing. Physica A: Statistical Mechanics and its Applications492 (2018), 138–154

2018
[63]

Daniele Notarmuzi, Claudio Castellano, Alessandro Flammini, Dario Mazzilli, and Filippo Radicchi. 2022. Universality, Criticality and Complexity of Information Propagation in Social Media.Nature Communications13, 1 (2022), 1308

2022
[64]

Alexandra Olteanu, Emre Kıcıman, and Carlos Castillo. 2018. A Critical Review of Online Social Data: Biases, Methodological Pitfalls, and Ethical Boundaries. InProceedings of the Eleventh ACM International Conference on Web Search and Data Mining(Marina Del Rey, CA, USA)(WSDM ’18). Association for Computing Machinery, New York, NY, USA, 785–786. doi:10.11...

work page doi:10.1145/3159652.3162004 2018
[65]

Scott E Page. 2012. Aggregation in agent-based models of economies.The Knowledge Engineering Review27, 2 (2012), 151–162

2012
[66]

Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative Agents: Interactive Simulacra of Human Behavior. InProceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. 1–22

2023
[67]

Zou, Aaron Shaw, Benjamin Mako Hill, Carrie Cai, Meredith Ringel Morris, Robb Willer, Percy Liang, and Michael S

Joon Sung Park, Carolyn Q. Zou, Aaron Shaw, Benjamin Mako Hill, Carrie Cai, Meredith Ringel Morris, Robb Willer, Percy Liang, and Michael S. Bernstein
[68]

LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals

Generative Agent Simulations of 1,000 People. arXiv:2411.10109 [cs.AI] https://arxiv.org/abs/2411.10109

work page internal anchor Pith review Pith/arXiv arXiv
[69]

Jinghua Piao, Yuwei Yan, Jun Zhang, Nian Li, Junbo Yan, Xiaochong Lan, Zhihong Lu, Zhiheng Zheng, Jing Yi Wang, Di Zhou, et al. 2025. Agentsociety: Large-scale simulation of llm-driven generative agents advances understanding of human behaviors and society.arXiv preprint arXiv:2502.08691(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[70]

William S Robinson. 2009. Ecological correlations and the behavior of individuals. International journal of epidemiology38, 2 (2009), 337–341

2009
[71]

Thomas C Schelling. 1971. Dynamic Models of Segregation.Journal of Mathe- matical Sociology1, 2 (1971), 143–186

1971
[72]

2006.Micromotives and Macrobehavior

Thomas C Schelling. 2006.Micromotives and Macrobehavior. WW Norton & Company

2006
[73]

Donald L Shaw, Maxwell McCombs, David H Weaver, and Bradley J Hamm
[74]

International journal of public opinion research11, 1 (1999), 2–24

Individuals, groups, and agenda melding: A theory of social dissonance. International journal of public opinion research11, 1 (1999), 2–24

1999
[75]

Zikai Song, Run Luo, Lintao Ma, Ying Tang, Yi-Ping Phoebe Chen, Junqing Yu, and Wei Yang. 2025. Temporal coherent object flow for multi-object tracking. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 6978–6986

2025
[76]

Flaminio Squazzoni. 2008. The Micro-Macro Link in Social Simulation.Sociologica 2, 1 (2008), 3–5

2008
[77]

Flaminio Squazzoni, Wander Jager, and Bruce Edmonds. 2014. Social Simulation in the Social Sciences: A Brief Overview.Social Science Computer Review32, 3 (2014), 279–294

2014
[78]

Kate Starbird, Ahmer Arif, and Tom Wilson. 2019. Disinformation as Collabora- tive Work: Surfacing the Participatory Nature of Strategic Information Operations. Proc. ACM Hum.-Comput. Interact.3, CSCW, Article 127 (Nov. 2019), 26 pages. doi:10.1145/3359229

work page doi:10.1145/3359229 2019
[79]

Melanie Subbiah, Amrita Bhattacharjee, Yilun Hua, Tharindu Kumarage, Huan Liu, and Kathleen McKeown. 2023. Towards Detecting Harmful Agendas in News Articles. InProceedings of the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, Jeremy Barnes, Orphée De Clercq, and Roman Klinger (Eds.). Association for Computa...

work page doi:10.18653/v1/2023.wassa-1.11 2023
[80]

Ziying Tan, Linbo Luo, Haiyan Yin, Yew-Soon Ong, and Wentong Cai. 2025. Crowd Dynamics Demand Adaptivity: Self-Adaptive Physics-Informed Neural Network for Crowd Simulation. InProceedings of the 33rd ACM International Conference on Multimedia(Dublin, Ireland)(MM ’25). Association for Computing Machinery, New York, NY, USA, 5913–5921. doi:10.1145/3746027.3754569

work page doi:10.1145/3746027.3754569 2025

Showing first 80 references.