At the inaugural AgentSociety Challenge hosted by The Web Conference 2025, a team from the Gaoling School of Artificial Intelligence at Renmin University of China (RUC), in collaboration with Huawei Poisson Lab, secured the silver medal (2nd place globally) in the User Modeling Track. The team, named JiuWen, was led by Professor Dou Zhicheng and consisted of PhD students Zhang Xinyu and Zhang Yuyao.

The AgentSociety Challenge is an international competition established by The Web Conference to advance the integration of AI and complex social systems. The competition explores the frontiers of AI agents in areas such as user behavior simulation and personalized recommendations. This year’s event attracted 295 teams from top universities, research institutions, and companies worldwide. After rigorous selection, 20 teams advanced to the finals in both the User Modeling Track and the Recommendation Track.

The award-winning work focuses on tackling key challenges in the automation of AI agent workflows, comprising three core components: an intelligent workflow construction engine, a prompt enhancement system, and a data-driven scheduling hub. Specifically, the team first developed a workflow search algorithm to dynamically decompose tasks and optimize execution paths, significantly improving multi-agent collaboration. Then, a self-optimizing prompt mechanism was designed to construct precise instruction chains, effectively mitigating error accumulation in agent collaboration and enhancing overall workflow performance. Finally, a dataset routing mechanism was implemented to optimize inference across multi-source heterogeneous data, boosting the system’s generalization capabilities in complex scenarios. The research findings have been accepted for presentation at The Web Conference 2025.

In the future, the team will continue to explore intelligent agent control technology to unlock more possibilities in the construction of multi-agent systems.