PersonaArena: Dynamic Simulation for Evaluating and Enhancing Persona-Level Role-Playing in Large Language Models
The article discusses the development of PersonaArena, a dynamic simulation framework aimed at enhancing persona-level role-playing in large language models (LLMs). It addresses the limitations of existing research, which often focuses on character-level settings and static evaluations. The framework utilizes a nuanced persona bank and features a multi-agent debating judge to improve the authenticity and social adeptness of AI agents.
- ▪PersonaArena is designed to evaluate and enhance persona-level role-playing in large language models.
- ▪The framework uses a large corpus of user-generated social content to create a detailed persona bank.
- ▪It includes a multi-agent debating judge for unbiased assessment of role-playing capabilities.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.17044 (cs) [Submitted on 16 May 2026] Title:PersonaArena: Dynamic Simulation for Evaluating and Enhancing Persona-Level Role-Playing in Large Language Models Authors:Wenlong Shi, Jianxun Lian, Mingqi Wu, Haiming Qin, Mingyang Zhou, Xing Xie, Naipeng Chao, Hao Liao View a PDF of the paper titled PersonaArena: Dynamic Simulation for Evaluating and Enhancing Persona-Level Role-Playing in Large Language Models, by Wenlong Shi and 7 other authors View PDF Abstract:Large language models (LLMs) increasingly serve as interactive social agents, yet their ability to maintain coherent and authentic persona-level role-playing remains limited, particularly in realistic social scenarios.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
