Elamur Research⌕
☰
home / papers / 2026 / GASim: A Graph-Accelerated Hybrid Framework for Social Simulation

GASim: A Graph-Accelerated Hybrid Framework for Social Simulation

Paper · 2026 · Xuan Zhou, Yanhui Sun, Hantao Yao et al.

Summary

Large-scale social simulators often pair a small number of LLM-driven "core" agents with a large population of rule-based agent-based model (ABM) agents, as in HiSim. These hybrid systems are still slow for two reasons. Core agents rely on LLM calls to retrieve memories, and ordinary agents are updated one at a time, so runtime grows linearly with population size.

GASim addresses both problems with graph methods:

  • Graph-Optimized Memory (GOM) stores each core agent's past neighbor messages as a sparse similarity graph weighted by opinion. It frames retrieval as an optimization task and solves it with cheap propagation, so no LLM is involved.
  • Graph Message Passing (GMP) replaces sequential ABM rules with a Graph Attention Network. It updates the opinions of all ordinary agents in one batched pass, using stance and profile features.
  • Entropy-Driven Grouping (EDG) chooses new core agents at every step. It picks agents whose neighborhoods hold the most varied opinions.

Compared with the traditional hybrid setup, GASim runs nearly ten times faster and uses far fewer tokens. It also follows real public opinion trends more closely, which makes large simulations practical at lower cost.

Key points

  • Efficiency: GASim is 9.94× faster end to end, with 16.39× speedup in the core-agent stage and 27.49× in the ordinary-agent stage. It uses less than 20% of baseline tokens. By the authors' estimate, running a 30-step simulation of 1M agents with sequential ABM updates would require more than 100 hours.
  • GOM graph design: each memory links to its top-k most similar memories. Edge weights multiply the two memories' opinion values (range −1 to +1) by the cosine similarity of their content embeddings.
  • LoCoMo benchmark: GOM reaches a state-of-the-art 71.56% accuracy, judged by LLM-as-a-Judge.
    • It gains about 10% on Single-Hop, Multi-Hop and Temporal questions.
    • It lags on Open-Domain questions, which make up about 5% of the data.
  • Ablation on the Politics dataset:
    • Replacing GOM with HiSim's LLM-based memory lowers Corr. from 0.4261 to 0.2942.
    • Replacing GMP with a Lorenz ABM performs worst (Corr. −0.0989).
    • Selecting core agents by degree instead of EDG raises ∆Div by 47.3%.
  • Code and limitations: code is at github.com/Jasmine0201/GASim. The authors note that synthetic opinion labels may carry LLM bias and that the simulation covers text only.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of Science and Technology of ChinaHefei · China145 works · 114,409 citations

Also on the byline: BASIS International.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksDecision Sciences · 249 worksManagement Science and Operations Research · 134 worksPhysics and Astronomy · 52 worksStatistical and Nonlinear Physics · 47 works

Cite

Cite as Xuan Zhou, Yanhui Sun, Hantao Yao, Allen He, Yongdong Zhang and Wu Liu. 2026. GASim: A Graph-Accelerated Hybrid Framework for Social Simulation. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.569

@inproceedings{zhou2026gasim,
  title = {{GASim: A Graph-Accelerated Hybrid Framework for Social Simulation}},
  author = {Xuan Zhou, Yanhui Sun, Hantao Yao, Allen He, Yongdong Zhang and Wu Liu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.569},
  url = {https://doi.org/10.18653/v1/2026.acl-long.569}
}

References

27 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shaked Brody, Uri Alon, and Eran Yahav. 2022. How attentive are graph attention networks? In The Tenth International Conference on Learning Representations, pages 1– 26. find
  2. Jinyuan Chen, Jiuchen Shi, Quan Chen, and Minyi Guo. 2025. Kairos: Low-latency multi-agent serving with shared llms and excessive loads in the public cloud. CoRR, abs/2508.06948. find
  3. Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, Yujia Qin, Xin Cong, Ruobing Xie, Zhiyuan Liu, Maosong Sun, and Jie Zhou. 2024. Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors. In The Twelft…
  4. Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. 2025. Mem0: Building production-ready AI agents with scalable long-term memory. CoRR, abs/2504.19413.
  5. Yun-Shiuan Chuang, Agam Goyal, Nikunj Harlalka, Siddharth Suresh, Robert Hawkins, Sijia Yang, Dhavan Shah, Junjie Hu, and Timothy T. Rogers. 2024. Simulating opinion dynamics with networks of llm-based agents. In Findings of the Association for Computational Linguistics, pages 3326–3346. find
  6. Guillaume Deffuant, Frédéric Amblard, Gérard Weis-buch, and Thierry Faure. 2002. How can extremism prevail? a study based on the relative agreement interaction model. Journal of Artificial Societies and Social Simulation, 5(04). find
  7. Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2024. The faiss library. CoRR, abs/2401.08281.
  8. Huiyu Min, Jiuxin Cao, Jiawei Ge, and Bo Liu. 2024. A multi-agent system for fine-grained opinion dynamics analysis in online social networks. IEEE Trans. Comput. Soc. Syst., 11(1):815–828. find
  9. Andrew B. Goldberg, Xiaojin Zhu, and Stephen Wright. 2007. Dissimilarity in graph-based semi-supervised classification. In Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics, pages 155–162. find
  10. Xinyi Mou, Zhongyu Wei, and Xuanjing Huang. 2024. Unveiling the truth and facilitating change: Towards agent-based large-scale social movement simulation. In Findings of the Association for Computational Linguistics, pages 4789–4809. find
  11. Rainer Hegselmann and Ulrich Krause. 2002. Opinion dynamics and bounded confidence models, analysis and simulation. Journal of Artificial Societies and Social Simulation, 5(03). find
  12. 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. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, page 1–22. find
  13. Zhiwei Jin, Juan Cao, Yongdong Zhang, and Jiebo Luo. 2016. News verification by exploiting conflicting social viewpoints in microblogs. In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, page 2972–2978. find
  14. Bai Jinbo and Li Hongbo. 2019. Study on a pareto principle case of social network. In Proceedings of the 2019 4th International Conference on Social Sciences and Economic Development, pages 113– 117. find
  15. Paul F. Lazarsfeld, Bernard Berelson, and Hazel Gaudet. 2021. The People’s Choice: How the Voter Makes Up His Mind in a Presidential Campaign. Columbia University Press. find
  16. Kun Liu, Qi Liu, Xinchen Liu, Jie Li, Yongdong Zhang, Jiebo Luo, Xiaodong He, and Wu Liu. 2025a. Hoigen-1m: A large-scale dataset for human-object interaction video generation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24001–24010. find
  17. Kun Liu, Mengxue Qu, Yang Liu, Yunchao Wei, Wenming Zhe, Yao Zhao, and Wu Liu. 2025b. Single-frame supervision for spatio-temporal video grounding. IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(7):5177–5191. find
  18. Yijun Liu, Wu Liu, Xiaoyan Gu, Yong Rui, Xiaodong He, and Yongdong Zhang. 2026. LMAgent: A large-scale multimodal agents society for multi-user simulation. IEEE Transactions on Multimedia, pages 1–12. find
  19. Jan Lorenz, Martin Neumann, and Tobias Schröder. 2021. Individual attitude change and societal dynamics: Computational experiments with psychological theories. Psychological Review, 128(04):623–642. find
  20. Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. 2024. Evaluating very long-term conversational memory of LLM agents. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, pages 13851–13870. find
  21. Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, and Daniel Chalef. 2025. Zep: A temporal knowledge graph architecture for agent memory. CoRR, abs/2501.13956.
  22. Yanhui Sun, Wu Liu, Wentao Wang, Hantao Yao, Jiebo Luo, and Yong-Dong Zhang. 2025. DynamiX: Large-scale dynamic social network simulator. CoRR, abs/2507.19929. — [arXiv:2507.19929](https://arxiv.org/abs/2507.19929) find
  23. Víctor Vargas-Pérez, Jesús Giráldez-Cru, Pablo Mesejo, and Oscar Cordón. 2025. Unveiling agents’ confidence in opinion dynamics models via graph neural networks. IEEE Trans. Comput. Soc. Syst., 12(2):725–737. find
  24. Kun Xiang, Zhili Liu, Terry Jingchen Zhang, Yinya Huang, Yunshuang Nie, Kaixin Cai, Yiyang Yin, Runhui Huang, Hanhui Li, Yihan Zeng, Yu-Jie Yuan, Jianhua Han, Lanqing Hong, Hang Xu, and Xiaodan Liang. 2026. AtomThink: Multimodal slow thinking with atomic step reasoning. IEEE Transactions on Pattern… find
  25. Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. 2025. A-Mem: Agentic memory for LLM agents. In The Thirty-ninth Annual Conference on Neural Information Processing Systems, pages 1–28.
  26. 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, and 4 others. 2024. OASIS: open agent social interaction simulati… — [arXiv:2411.11581](https://arxiv.org/abs/2411.11581) find
  27. Jun Zhang, Yuwei Yan, Junbo Yan, Zhiheng Zheng, Jinghua Piao, Depeng Jin, and Yong Li. 2025. A parallelized framework for simulating large-scale LLM agents with realistic environments and interactions. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2026-gasim-graph-accelerated-hybrid-framework/