arxiv: 2604.18011 · v1 · submitted 2026-04-20 · 💻 cs.SI · cs.DB

Recognition: unknown

Topology-Aware LLM-Driven Social Simulation: A Unified Framework for Efficient and Realistic Agent Dynamics

Yuwei Xu , Shulun Zhang , Yingli Zhou , Shipei Zeng , Laks V.S. Lakshmanan , Chenhao Ma

Authors on Pith no claims yet

Pith reviewed 2026-05-10 03:42 UTC · model grok-4.3

classification 💻 cs.SI cs.DB

keywords social simulationlarge language modelsnetwork topologymulti-agent systemsagent-based modelingstructural influencesimulation efficiency

0 comments

The pith

TopoSim integrates network topology into LLM agent simulations to cut token use by 50-90% while improving fidelity to real social structures.

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

The paper proposes TopoSim, a framework that makes LLM-driven social simulations both more efficient and more realistic by explicitly using the underlying network structure. Existing approaches treat networks as fixed scaffolds without leveraging how topology shapes influence and behavior, leading to wasteful computation and unrealistic dynamics. TopoSim groups agents with similar structural roles into shared backbone units for coordinated updates and models influence as varying according to network connections rather than uniformly. Experiments show this reduces token consumption dramatically while matching or exceeding previous fidelity and better capturing phenomena like structural patterns in real systems. A sympathetic reader would care because efficient, large-scale social simulations could help study collective behavior without enormous computational costs.

Core claim

TopoSim is a unified topology-aware social simulation framework that aligns agents into backbone units based on structural roles and models social influence as a structure-induced signal. This allows coordinated agent updates that preserve emergent dynamics while cutting redundant LLM calls, and introduces heterogeneous interaction patterns grounded in network topology instead of uniform assumptions.

What carries the argument

Backbone units that group agents with similar structural roles and interaction contexts for coordinated updates, together with topology-derived heterogeneous influence patterns.

If this is right

Token consumption in LLM-based simulations drops by 50 to 90 percent.
Simulations more accurately reproduce structural phenomena from real-world social systems.
The approach generalizes across different simulation frameworks and scales well.
Emergent social dynamics remain preserved despite the efficiency optimizations.

Where Pith is reading between the lines

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

Similar topology-aware grouping might reduce costs in other LLM multi-agent applications such as collaborative problem solving.
Large population simulations become feasible without linear increases in compute resources.
Future work could test whether these structural signals improve predictive accuracy for real events modeled in simulations.

Load-bearing premise

That grouping agents by structural roles into backbone units and deriving influence from topology will preserve emergent social dynamics without introducing grouping artifacts or losing LLM reasoning quality.

What would settle it

Compare simulations of the same scenarios run with and without the backbone unit grouping: if the version without grouping achieves equal or better fidelity and structural accuracy at similar or lower token cost, the benefit of the topology integration would be called into question.

Figures

Figures reproduced from arXiv: 2604.18011 by Chenhao Ma, Laks V.S. Lakshmanan, Shipei Zeng, Shulun Zhang, Yingli Zhou, Yuwei Xu.

**Figure 3.** Figure 3: Given an input social graph and the initial state of each [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 3.** Figure 3: Coordinated LLM updates (top) and emergent role differentiation (bottom) in TopoSim. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Impact of Update Coordination on Social Dynamics Across ECS and OASIS. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Node-level consistency diagnostics on ECS. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 7.** Figure 7: Role Differentiation Effect and Ablation Study on [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 8.** Figure 8: Cross-LLM robustness and scalability on ECS. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗

read the original abstract

Social simulation is essential for understanding collective human behavior by modeling how individual interactions give rise to large-scale social dynamics. Recent advances in large language models (LLMs) have enabled multi-agent frameworks with human-like reasoning and communication capabilities. However, existing LLM-based simulations treat social networks as fixed communication scaffolds, failing to leverage the structural signals that shape behavioral convergence and heterogeneous influence in real-world systems, which often leads to inefficient and unrealistic dynamics. To address this challenge, we propose TopoSim, a unified topology-aware social simulation framework that explicitly integrates structural reasoning into agent interactions along two complementary dimensions. First, TopoSim aligns agents with similar structural roles and interaction contexts into shared backbone units, enabling coordinated updates that reduce redundant computation while preserving emergent social dynamics. Second, TopoSim models social influence as a structure-induced signal, introducing heterogeneous interaction patterns grounded in network topology rather than uniform influence assumptions. Extensive experiments across three social simulation frameworks and diverse datasets demonstrate that TopoSim achieves comparable or improved simulation fidelity while reducing token consumption by 50 - 90%. Moreover, our approach more accurately reproduces key structural phenomena observed in real-world social systems and exhibits strong generalization and scalability.

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

The paper shows how to cut token costs in LLM social simulations by half or more through topology-based agent grouping, though this risks altering the very dynamics it aims to model more accurately.

read the letter

I looked over the TopoSim paper. The main point is that they have a framework to make large-scale LLM agent social simulations run with far less token cost by using the network topology to organize the agents. What is new is the idea of backbone units that group agents with similar structural positions for coordinated LLM updates, paired with injecting topology-based signals to create varied influence patterns. This moves beyond treating the network as just a static graph for who talks to whom. The experiments cover three different simulation setups and multiple datasets, with results showing token reductions in the 50 to 90 percent range and closer matches to real social network properties like community structure or influence spread. The work does a solid job on the efficiency side and provides evidence that the topology integration helps with realism. The scalability claims seem grounded in the tests they ran. The potential issue is whether the backbone grouping keeps enough individual variation. When agents in a unit update together, their LLM responses might get pulled toward a common path, which could dampen the stochastic differences that make social dynamics interesting and realistic. The paper asserts that emergent behaviors are preserved, but without detailed ablations on variance or comparisons to fully independent runs, it's possible the improved structural fidelity comes partly from this synchronization rather than the topology signals alone. That assumption about no artifacts needs checking. Readers who work on agent-based modeling with LLMs or who study collective behavior computationally would find this relevant. It gives a concrete way to handle bigger populations without proportional compute increases. The paper has enough new ideas and reported results to go through peer review, where the focus should be on verifying the experimental design against the heterogeneity concern. I would recommend putting it in front of referees.

Referee Report

2 major / 2 minor

Summary. The paper proposes TopoSim, a unified topology-aware framework for LLM-driven social simulations. It integrates network structure in two ways: (1) grouping agents with similar structural roles and interaction contexts into shared backbone units to enable coordinated updates that reduce redundant LLM calls, and (2) modeling social influence as a heterogeneous, topology-induced signal rather than uniform. The central claims are that this yields 50-90% token savings while achieving comparable or superior simulation fidelity, more accurately reproducing real-world structural phenomena (e.g., influence propagation and convergence patterns), and exhibiting strong generalization and scalability across three existing simulation frameworks and diverse datasets.

Significance. If the efficiency and fidelity claims are substantiated, the work would be a meaningful contribution to multi-agent LLM social simulation by showing how explicit use of network topology can simultaneously improve computational efficiency and realism. The unified framework approach and the explicit separation of structural role alignment from influence modeling are conceptually clean and address a recognized limitation in current fixed-scaffold agent models. No machine-checked proofs or parameter-free derivations are present, but the emphasis on reproducible structural metrics and cross-framework testing is a positive direction.

major comments (2)

[Abstract and §4] Abstract and §4 (Experiments): The manuscript asserts 'extensive experiments across three social simulation frameworks and diverse datasets' that deliver 50-90% token savings and improved structural fidelity, yet supplies no quantitative results, tables, figures, baseline comparisons, control conditions, or statistical tests. Without these, the central efficiency and fidelity claims cannot be evaluated and remain load-bearing assertions rather than demonstrated outcomes.
[§3.1] §3.1 (Backbone units): The coordinated-update mechanism for agents grouped by structural role is presented as preserving emergent dynamics. However, the description does not specify how per-agent LLM reasoning paths, stochasticity, and state variance are maintained once updates are coordinated within a unit. This directly risks the secondary claim of improved reproduction of structural phenomena, because any implicit averaging or synchronization would alter heterogeneity and influence propagation even if topology signals are added.

minor comments (2)

[§3] Notation for 'backbone units' and 'structure-induced influence' is introduced without a compact mathematical definition or pseudocode; a short formalization would improve clarity and reproducibility.
[Abstract and §1] The abstract and introduction would benefit from explicit citation of prior topology-aware social simulation work (e.g., network influence models outside LLM agents) to better situate the novelty.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We address the two major comments point by point below and will revise the manuscript to strengthen the presentation of results and clarify the technical mechanisms.

read point-by-point responses

Referee: [Abstract and §4] Abstract and §4 (Experiments): The manuscript asserts 'extensive experiments across three social simulation frameworks and diverse datasets' that deliver 50-90% token savings and improved structural fidelity, yet supplies no quantitative results, tables, figures, baseline comparisons, control conditions, or statistical tests. Without these, the central efficiency and fidelity claims cannot be evaluated and remain load-bearing assertions rather than demonstrated outcomes.

Authors: We appreciate the referee highlighting the need for clearer substantiation of the empirical claims. The manuscript does describe the experimental protocol, datasets, and high-level outcomes in §4, including cross-framework testing. However, to make the quantitative evidence fully transparent and evaluable, we will revise §4 to include explicit tables reporting token consumption reductions (with per-framework breakdowns), fidelity metrics against real-world structural benchmarks, baseline comparisons, control conditions, and statistical significance tests. These additions will directly support the 50-90% savings and fidelity claims. revision: yes
Referee: [§3.1] §3.1 (Backbone units): The coordinated-update mechanism for agents grouped by structural role is presented as preserving emergent dynamics. However, the description does not specify how per-agent LLM reasoning paths, stochasticity, and state variance are maintained once updates are coordinated within a unit. This directly risks the secondary claim of improved reproduction of structural phenomena, because any implicit averaging or synchronization would alter heterogeneity and influence propagation even if topology signals are added.

Authors: We thank the referee for this important clarification request. In the backbone unit design, coordination applies only to update scheduling and the injection of shared structural context (e.g., role embeddings and topology signals); each agent continues to perform its own independent LLM inference call with its private state, full reasoning trace, and independent stochastic sampling (temperature, top-p, etc.). No state averaging or output synchronization occurs within a unit. This separation ensures that heterogeneity and influence propagation remain intact. We will expand §3.1 with a precise description of this per-agent independence, including a diagram or pseudocode example, to eliminate any ambiguity. revision: yes

Circularity Check

0 steps flagged

No circularity in framework proposal or experimental claims

full rationale

The paper introduces TopoSim as a novel framework with two explicit design components—backbone units for role-aligned coordinated updates and topology-induced heterogeneous influence—presented as solutions to limitations in prior LLM simulations. These are architectural choices, not mathematical derivations. No equations, parameter fittings, or self-citation chains are described that reduce any claimed result (e.g., 50-90% token reduction or improved fidelity) to the inputs by construction. Performance assertions rest on experimental validation across multiple frameworks and datasets, rendering the central claims independent and self-contained rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are detailed. The framework introduces 'backbone units' and 'structure-induced signals' whose precise definitions and assumptions remain unspecified.

pith-pipeline@v0.9.0 · 5528 in / 1023 out tokens · 25767 ms · 2026-05-10T03:42:00.728739+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

38 extracted references · 22 canonical work pages · 1 internal anchor

[1]

Sinan Aral and Dylan Walker. 2012. Identifying influential and susceptible members of social networks.Science337, 6092 (2012), 337–341

2012
[2]

and Mason, Winter A

Eytan Bakshy, Jake M. Hofman, Winter A. Mason, and Duncan J. Watts. 2011. Everyone’s an influencer: quantifying influence on twitter. InProceedings of the Fourth ACM International Conference on Web Search and Data Mining(Hong Kong, China)(WSDM ’11). Association for Computing Machinery, New York, NY, USA, 65–74. https://doi.org/10.1145/1935826.1935845

work page doi:10.1145/1935826.1935845 2011
[3]

Garrett Bernstein and Kyle O’Brien. 2013. Stochastic agent-based simulations of social networks. InProceedings of the 46th Annual Simulation Symposium(San Diego, California)(ANSS 13). Society for Computer Simulation International, San Diego, CA, USA, Article 5, 8 pages

2013
[4]

Anwoy Chatterjee, H S V N S Kowndinya Renduchintala, Sumit Bhatia, and Tanmoy Chakraborty. 2024. POSIX: A Prompt Sensitivity Index For Large Language Models. InFindings of the Association for Computational Linguistics: EMNLP 2024, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen (Eds.). As- sociation for Computational Linguistics, Miami, Florida, USA, 14...

work page doi:10.18653/v1/2024.findings-emnlp.852 2024
[6]

Yun-Shiuan Chuang, Agam Goyal, Nikunj Harlalka, Siddharth Suresh, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, and Timothy Rogers. 2024. Simulat- ing Opinion Dynamics with Networks of LLM-based Agents. InFindings of the As- sociation for Computational Linguistics: NAACL 2024, Kevin Duh, Helena Gomez, and Steven Bethard (Eds.). Association for Compu...

work page doi:10.18653/v1/2024.findings-naacl.211 2024
[7]

Guillaume Deffuant, David Neau, Frédéric Amblard, and Gérard Weisbuch. 2000. Mixing beliefs among interacting agents.Adv. Complex Syst.3 (2000), 87–98. https://api.semanticscholar.org/CorpusID:15604530

2000
[8]

Claire Donnat and Susan Holmes. 2018. Tracking network dynamics: A survey using graph distances.The Annals of Applied Statistics12, 2 (2018), 971–1012

2018
[9]

Wellman, Susan Carey, Lila Gleitman, Elissa L

Joshua M. Epstein and Robert L. Axtell. 1996.Growing Artificial Societies: Social Science from the Bottom Up. The MIT Press. https://doi.org/10.7551/mitpress/ 3374.001.0001

work page doi:10.7551/mitpress/ 1996
[10]

Stephen Eubank, Hasan Guclu, V. S. Anil Kumar, Madhav V. Marathe, Aravind Srinivasan, Zoltán Toroczkai, and Nan Wang. 2004. Modelling disease outbreaks in realistic urban social networks.Nature429, 6988 (01 May 2004), 180–184. https://doi.org/10.1038/nature02541

work page doi:10.1038/nature02541 2004
[11]

Ferguson, Derek A.T

Neil M. Ferguson, Derek A.T. Cummings, Simon Cauchemez, Christophe Fraser, Steven Riley, Aronrag Meeyai, Sopon Iamsirithaworn, and Donald S. Burke. 2005. Strategies for containing an emerging influenza pandemic in Southeast Asia. Nature437, 7056 (01 Sep 2005), 209–214. https://doi.org/10.1038/nature04017

work page doi:10.1038/nature04017 2005
[12]

Antonino Ferraro, Antonio Galli, Valerio La Gatta, Marco Postiglione, Gian Marco Orlando, Diego Russo, Giuseppe Riccio, Antonio Romano, and Vincenzo Moscato
[13]

InSocial Networks Analysis and Mining, Luca Maria Aiello, Tanmoy Chakraborty, and Sabrina Gaito (Eds.)

Agent-Based Modelling Meets Generative AI in Social Network Simulations. InSocial Networks Analysis and Mining, Luca Maria Aiello, Tanmoy Chakraborty, and Sabrina Gaito (Eds.). Springer Nature Switzerland, Cham, 155–170
[14]

Noah Friedkin and Eugene Johnsen. 1990. Social Influence and Opinions.Journal of Mathematical Sociology - J MATH SOCIOL15 (01 1990), 193–206. https: //doi.org/10.1080/0022250X.1990.9990069

work page doi:10.1080/0022250x.1990.9990069 1990
[15]

Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, and Yong Li. 2023. S3: Social-network simulation system with large language model-empowered agents.arXiv preprint arXiv:2307.14984 (2023)

work page arXiv 2023
[16]

Chen Gao, Fengli Xu, Xu Chen, Xiang Wang, Xiangnan He, and Yong Li. 2024. Simulating Human Society with Large Language Model Agents: City, Social Me- dia, and Economic System. InCompanion Proceedings of the ACM Web Conference 2024(Singapore, Singapore)(WWW ’24). Association for Computing Machinery, New York, NY, USA, 1290–1293. https://doi.org/10.1145/358...

work page doi:10.1145/3589335.3641253 2024
[17]

Rainer Hegselmann and Ulrich Krause. 2002. Opinion Dynamics and Bounded Confidence Models, Analysis and Simulation.Journal of Artificial Societies and Social Simulation5 (07 2002)

2002
[18]

Keith Henderson, Brian Gallagher, Tina Eliassi-Rad, Hanghang Tong, Sugato Basu, Leman Akoglu, Danai Koutra, Christos Faloutsos, and Lei Li. 2012. RolX: structural role extraction & mining in large graphs. InProceedings of the 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (Beijing, China)(KDD ’12). Association for Computin...

work page doi:10.1145/2339530.2339723 2012
[19]

Guiyang Hou, Wenqi Zhang, Zhe Zheng, Yongliang Shen, and Weiming Lu
[20]

In: NAACL (Long Papers)

Scaling LLMs’ Social Reasoning: Sprinkle Cognitive “Aha Moment” into Fundamental Long-thought Logical Capabilities. InFindings of the Association for Computational Linguistics: ACL 2025, Wanxiang Che, Joyce Nabende, Ekate- rina Shutova, and Mohammad Taher Pilehvar (Eds.). Association for Computa- tional Linguistics, Vienna, Austria, 3126–3138. https://doi...

work page doi:10.18653/v1/2025 2025
[21]

Glen Jeh and Jennifer Widom. 2003. Scaling personalized web search. InProceed- ings of the 12th international conference on World Wide Web. 271–279

2003
[22]

Zexun Jiang, Yafang Shi, Maoxu Li, Hongjiang Xiao, Yunxiao Qin, Qinglan Wei, Ye Wang, and Yuan Zhang. 2024. Casevo: A Cognitive Agents and Social Evolution Simulator. arXiv:2412.19498 [cs.SI] https://arxiv.org/abs/2412.19498

work page arXiv 2024
[23]

Unchitta Kan, Michelle Feng, and Mason A Porter. 2023. An adaptive bounded- confidence model of opinion dynamics on networks.Journal of Complex Networks 11, 1 (2023), 415–444. https://doi.org/10.1093/comnet/cnac055

work page doi:10.1093/comnet/cnac055 2023
[24]

Yuhan Liu, Zirui Song, Juntian Zhang, Xiaoqing Zhang, Xiuying Chen, and Rui Yan. 2025. The Stepwise Deception: Simulating the Evolution from True News to Fake News with LLM Agents. arXiv:2410.19064 [cs.SI] https://arxiv.org/abs/ 2410.19064

work page arXiv 2025
[25]

Francois Lorrain and Harrison C White. 1971. Structural equivalence of indi- viduals in social networks.The Journal of mathematical sociology1, 1 (1971), 49–80

1971
[26]

FRANÇOIS LORRAIN and HARRISON C. WHITE. 1977. STRUCTURAL EQUIV- ALENCE OF INDIVIDUALS IN SOCIAL NETWORKS. InSocial Networks, Samuel Leinhardt (Ed.). Academic Press, 67–98. https://doi.org/10.1016/B978-0-12- 442450-0.50012-2

work page doi:10.1016/b978-0-12- 1977
[27]

1999.The PageRank citation ranking: Bringing order to the web.Technical Report

Lawrence Page, Sergey Brin, Rajeev Motwani, and Terry Winograd. 1999.The PageRank citation ranking: Bringing order to the web.Technical Report. Stanford infolab

1999
[28]

Bernstein

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(San Francisco, CA, USA)(UIST ’23). Association for Computing Machinery, New York, NY, USA, ...

work page doi:10.1145/3586183.3606763 2023
[29]

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
[30]

Ribeiro, Pedro H.P

Leonardo F.R. Ribeiro, Pedro H.P. Saverese, and Daniel R. Figueiredo. 2017. struc2vec: Learning Node Representations from Structural Identity. InProceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’17). ACM, 385–394. https://doi.org/10.1145/3097983.3098061

work page doi:10.1145/3097983.3098061 2017
[31]

Chenxi Wang, Zongfang Liu, Dequan Yang, and Xiuying Chen. 2025. Decoding Echo Chambers: LLM-Powered Simulations Revealing Polarization in Social Networks. InProceedings of the 31st International Conference on Computational Linguistics, Owen Rambow, Leo Wanner, Marianna Apidianaki, Hend Al-Khalifa, Barbara Di Eugenio, and Steven Schockaert (Eds.). Associat...

2025
[32]

Watts and Peter Sheridan Dodds

Duncan J. Watts and Peter Sheridan Dodds. 2007. Influentials, Networks, and Public Opinion Formation.Journal of Consumer Research34, 4 (05 2007), 441–

2007
[33]

https://doi.org/10.1086/518527 arXiv:https://academic.oup.com/jcr/article- pdf/34/4/441/17929486/34-4-441.pdf

work page doi:10.1086/518527
[34]

Watts, Duncan J.and Strogatz

Steven H. Watts, Duncan J.and Strogatz. 1998. Collective dynamics of ‘small- world’ networks.Nature393, 6684 (01 Jun 1998), 440–442. https://doi.org/10. 1038/30918

1998
[35]

Harrison C White, Scott A Boorman, and Ronald L Breiger. 1976. Social structure from multiple networks. I. Blockmodels of roles and positions.American journal of sociology81, 4 (1976), 730–780

1976
[36]

Ziyi Yang, Zaibin Zhang, Zirui Zheng, Yuxian Jiang, Ziyue Gan, Zhiyu Wang, Zijian Ling, Jinsong Chen, Martz Ma, Bowen Dong, Prateek Gupta, Shuyue Hu, Zhenfei Yin, Guohao Li, Xu Jia, Lijun Wang, Bernard Ghanem, Huchuan Lu, Chaochao Lu, Wanli Ouyang, Yu Qiao, Philip Torr, and Jing Shao. 2025. OASIS: Open Agent Social Interaction Simulations with One Million...

work page arXiv 2025
[37]

Junchi Yao, Hongjie Zhang, Jie Ou, Dingyi Zuo, Zheng Yang, and Zhicheng Dong. 2025. Social opinions prediction utilizes fusing dynamics equation with LLM-based agents.Scientific Reports15, 1 (03 May 2025), 15472. https://doi.org/ 10.1038/s41598-025-99704-3

work page doi:10.1038/s41598-025-99704-3 2025
[38]

Haofei Yu, Zhaochen Hong, Zirui Cheng, Kunlun Zhu, Keyang Xuan, Jinwei Yao, Tao Feng, and Jiaxuan You. 2025. ResearchTown: Simulator of Human Research Community. arXiv:2412.17767 [cs.CL] https://arxiv.org/abs/2412.17767

work page arXiv 2025
[39]

Albert Ĺaszló Barabási and Réka Albert. 1999. Emergence of scaling in random networks.Science286 5439 (1999), 509–12. https://api.semanticscholar.org/ CorpusID:260558609

1999