Wenjun Liu


Fixing paper assignments

  1. Please select all papers that belong to the same person.
  2. Indicate below which author they should be assigned to.
Provide a valid ORCID iD here. This will be used to match future papers to this author.
Provide the name of the school or the university where the author has received or will receive their highest degree (e.g., Ph.D. institution for researchers, or current affiliation for students). This will be used to form the new author page ID, if needed.

TODO: "submit" and "cancel" buttons here


2025

pdf bib
Judging with Many Minds: Do More Perspectives Mean Less Prejudice? On Bias Amplification and Resistance in Multi-Agent Based LLM-as-Judge
Chiyu Ma | Enpei Zhang | Yilun Zhao | Wenjun Liu | Yaning Jia | Peijun Qing | Lin Shi | Arman Cohan | Yujun Yan | Soroush Vosoughi
Findings of the Association for Computational Linguistics: EMNLP 2025

LLM-as-Judge has emerged as a scalable alternative to human evaluation, enabling large language models (LLMs) to provide reward signals in trainings. While recent work has explored multi-agent extensions such as multi-agent debate and meta-judging to enhance evaluation quality, the question of how intrinsic biases manifest in these settings remains underexplored. In this study, we conduct a systematic analysis of four diverse bias types: position bias, verbosity bias, chain-of-thought bias, and bandwagon bias. We evaluate these biases across two widely adopted multi-agent LLM-as-Judge frameworks: Multi-Agent-Debate and LLM-as-Meta-Judge. Our results show that debate framework amplifies biases sharply after the initial debate, and this increased bias is sustained in subsequent rounds, while meta-judge approaches exhibit greater resistance. We further investigate the incorporation of PINE, a leading single-agent debiasing method, as a bias-free agent within these systems. The results reveal that this bias-free agent effectively reduces biases in debate settings but provides less benefit in meta-judge scenarios. Our work provides a comprehensive study of bias behavior in multi-agent LLM-as-Judge systems and highlights the need for targeted bias mitigation strategies in collaborative evaluation settings.