OffsetBias: Leveraging Debiased Data for Tuning Evaluators

Junsoo Park, Seungyeon Jwa, Ren Meiying, Daeyoung Kim, Sanghyuk Choi


Abstract
Employing Large Language Models (LLMs) to assess the quality of generated responses has become a widely adopted evaluation method. Specifically, instruct-tuned models and fine-tuned judge models based on open-source LLMs have been reported. While it is known that judge models are vulnerable to certain biases, such as favoring longer answers regardless of content, the specifics of these biases remain under-explored. In this work, we qualitatively identify six types of biases inherent in various judge models. We propose EvalBiasBench as a meta-evaluation collection of hand-crafted test cases for each bias type. Additionally, we present de-biasing dataset construction methods and the associated preference dataset OffsetBias. Experimental results demonstrate that fine-tuning on our dataset significantly enhances the robustness of judge models against biases and improves performance across most evaluation scenarios. We release our datasets and the fine-tuned judge model to public.
Anthology ID:
2024.findings-emnlp.57
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2024
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
1043–1067
Language:
URL:
https://aclanthology.org/2024.findings-emnlp.57
DOI:
Bibkey:
Cite (ACL):
Junsoo Park, Seungyeon Jwa, Ren Meiying, Daeyoung Kim, and Sanghyuk Choi. 2024. OffsetBias: Leveraging Debiased Data for Tuning Evaluators. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1043–1067, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
OffsetBias: Leveraging Debiased Data for Tuning Evaluators (Park et al., Findings 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.findings-emnlp.57.pdf
Data:
 2024.findings-emnlp.57.data.zip