让AI通过类比推理更准确判断模糊内容，避免误判。
CARO: Chain-of-Analogy Reasoning Optimization for Robust Content Moderation
- 用类比推理链引导AI思考，避免依赖表面线索做判断。
- 在复杂模糊内容上提升24.9%的识别准确率，优于顶尖模型。
- 适合需要高可靠性内容审核的平台或安全研究者使用。
当前大型语言模型即使经过专门推理训练，仍因上下文中的误导性‘决策捷径’而在模糊内容审核中表现不佳。受认知心理学中专家审核机制启发，我们提出 exttt{CARO}（类比推理优化）——一种两阶段训练框架，以诱导大模型具备稳健的类比推理能力。首先，基于审核数据通过检索增强生成（RAG）构建类比推理链，并进行监督微调（SFT）；其次，设计定制化的直接偏好优化（DPO）方法，显式强化类比推理行为。与静态检索不同， exttt{CARO} 在推理时动态生成定制类比参考，有效缓解有害决策捷径。大量实验表明， exttt{CARO} 显著优于现有先进推理模型（DeepSeek R1、QwQ）、专用审核模型（LLaMA Guard）及前沿微调与检索增强方法，在挑战性模糊审核基准上平均F1分数提升24.9%。
原文摘要 · Abstract (English)
Current large language models (LLMs), even those explicitly trained for reasoning, often struggle with ambiguous content moderation cases due to misleading "decision shortcuts" embedded in context. Inspired by cognitive psychology insights into expert moderation, we introduce \caro (Chain-of-Analogy Reasoning Optimization), a novel two-stage training framework to induce robust analogical reasoning in LLMs. First, \caro bootstraps analogical reasoning chains via retrieval-augmented generation (RAG) on moderation data and performs supervised fine-tuning (SFT). Second, we propose a customized direct preference optimization (DPO) approach to reinforce analogical reasoning behaviors explicitly. Unlike static retrieval methods, \caro dynamically generates tailored analogical references during inference, effectively mitigating harmful decision shortcuts. Extensive experiments demonstrate that \caro substantially outperforms state-of-the-art reasoning models (DeepSeek R1, QwQ), specialized moderation models (LLaMA Guard), and advanced fine-tuning and retrieval-augmented methods, achieving an average F1 score improvement of 24.9\% on challenging ambiguous moderation benchmarks.
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。