构建开放问答场景下的新基准，有效检测并减轻大模型社会偏见。
Evaluating and Mitigating Social Bias for Large Language Models in Open-ended Settings
- 扩展原数据集，新增填空与简答两类开放题型。
- 提出复合提示法，在不降低准确率前提下显著减少偏见。
- 适合关注大模型公平性与安全性的研究者与开发者。
当前大语言模型的社会偏见评估多依赖预设题型（如选择题），难以反映真实交互的开放性。为此，我们扩展了现有数据集 BBQ 为 Open-BBQ，新增填空与简答两类开放题型，构建面向开放问答场景的全面评估框架。针对开放文本中偏见难以量化的问题，我们设计了基于句子与段落标注的评估流程。实验发现，现有去偏方法（如 self-debiasing）存在过度修正问题，导致正确答案被错误修改。为此，我们提出 Composite Prompting——一种结合结构化示例与显式思维链的上下文学习方法，形成统一指令模板，使模型能主动识别需去偏内容。在 GPT-3.5 与 GPT-4o 上的实验表明，该方法显著降低偏见水平，同时保持高准确率。
原文摘要 · Abstract (English)
Current social bias benchmarks for Large Language Models (LLMs) primarily rely on predefined question formats like multiple-choice, limiting their ability to reflect the complexity and open-ended nature of real-world interactions. To close this gap, we extend an existing dataset BBQ (Parrish et al., 2022) to Open-BBQ, a comprehensive framework to evaluate the social bias of LLMs in open-ended settings by incorporating two additional question categories: fill-in-the-blank and short-answer. Since our new Open-BBQ dataset contains a lot of open-ended responses like sentences and paragraphs, we developed an evaluation process to detect biases from open-ended content by labeling sentences and paragraphs. In addition to this, we also found that existing debiasing methods, such as self-debiasing (Gallegos et al., 2024), have over-correction issues, which make the original correct answers incorrect. In order to solve this issue, we propose Composite Prompting, an In-context Learning (ICL) method combining structured examples with explicit chain-of-thought reasoning to form a unified instruction template for LLMs to explicitly identify content that needs debiasing. Experimental results show that the proposed method significantly reduces the bias for both GPT-3.5 and GPT-4o while maintaining high accuracy.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。