Skip to content

About

Project

SocioVerse is a longitudinal, LLM-native social-simulation runtime built around Lewin's field theory, B = f(P, E).

It is developed by Shanghai Innovation Institute together with Fudan University.

A hosted instance runs at socioverse.fudan-disc.com — see SocioVerse Online.

License

SocioVerse is released under the Apache License 2.0. The full text ships as LICENSE in the repository.

Citation

If SocioVerse2 is useful in your research, please cite the technical report (arXiv:2609.24911):

Xinnong Zhang, Jiayu Lin, Jia Wang, et al. SocioVerse2: A Longitudinal Dynamic Social Simulation Framework under a Human-AI Co-evolutionary Paradigm. arXiv:2609.24911, 2026.

@misc{zhang2026socioverse2,
  title         = {SocioVerse2: A Longitudinal Dynamic Social Simulation Framework under a Human-AI Co-evolutionary Paradigm},
  author        = {Xinnong Zhang and
                   Jiayu Lin and
                   Jia Wang and
                   Yixu Huang and
                   Xinyi Mou and
                   Yingqian Wu and
                   Jingcong Liang and
                   Shijun Lei and
                   Jianing Shi and
                   Guanying Li and
                   Siyuan Wang and
                   Hanjia Lyu and
                   Zhenfei Yin and
                   Yunlu Yin and
                   Siming Chen and
                   Yulan He and
                   Jiebo Luo and
                   Xuanjing Huang and
                   Liyin Jin and
                   Baohua Zhou and
                   Hanqi Yan and
                   Zhongyu Wei},
  year          = {2026},
  eprint        = {2609.24911},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2609.24911}
}

Contributing

Bug reports and study contributions are welcome on the issue tracker; read CONTRIBUTING.md first, and report security issues as SECURITY.md describes. Contributed studies follow the standard anatomy — four interface implementations, validated artifacts, a grounding ledger — and join the routable catalog like any bundled study. See Anatomy of a Study and Build a Study from Scratch.

Acknowledgements

Several of the bundled reference studies stand on existing work, and each records its debts in its own grounding ledger rather than only here:

  • Chicago segregation builds on Schelling's segregation model and on US Census tract data.
  • Opinion dynamics builds on HiSim (Mou et al., ACL Findings 2024) and on the Deffuant bounded-confidence model.
  • The ABM benchmark family mirrors eleven classic agent-based models, each checked for parity against its original implementation.

See Examples for what each one demonstrates.