Mitigating the Bias of Large Language Model Evaluation

Zhou Hongli, Huang Hui, Long Yunfei, Xu Bing, Zhu Conghui, Cao Hailong, Yang Muyun, Zhao Tiejun


Abstract
“Recently, there has been a trend of evaluating the Large Language Model (LLM) quality in theflavor of LLM-as-a-Judge, namely leveraging another LLM to evaluate the current output qual-ity. However, existing judges are proven to be biased, namely they would favor answers whichpresent better superficial quality (such as verbosity, fluency) while ignoring the instruction fol-lowing ability. In this work, we propose systematic research about the bias of LLM-as-a-Judge.Specifically, for closed-source judge models, we apply calibration to mitigate the significance ofsuperficial quality, both on probability level and prompt level. For open-source judge models, wepropose to mitigate the bias by contrastive training, with curated negative samples that deviatefrom instruction but present better superficial quality. We apply our methods on the bias evalu-ation benchmark, and experiment results show our methods mitigate the bias by a large marginwhile maintaining a satisfactory evaluation accuracy.”
Anthology ID:
2024.ccl-1.101
Volume:
Proceedings of the 23rd Chinese National Conference on Computational Linguistics (Volume 1: Main Conference)
Month:
July
Year:
2024
Address:
Taiyuan, China
Editors:
Maosong Sun, Jiye Liang, Xianpei Han, Zhiyuan Liu, Yulan He
Venue:
CCL
SIG:
Publisher:
Chinese Information Processing Society of China
Note:
Pages:
1310–1319
Language:
English
URL:
https://aclanthology.org/2024.ccl-1.101/
DOI:
Bibkey:
Cite (ACL):
Zhou Hongli, Huang Hui, Long Yunfei, Xu Bing, Zhu Conghui, Cao Hailong, Yang Muyun, and Zhao Tiejun. 2024. Mitigating the Bias of Large Language Model Evaluation. In Proceedings of the 23rd Chinese National Conference on Computational Linguistics (Volume 1: Main Conference), pages 1310–1319, Taiyuan, China. Chinese Information Processing Society of China.
Cite (Informal):
Mitigating the Bias of Large Language Model Evaluation (Hongli et al., CCL 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.ccl-1.101.pdf