CriticGen：将生成感知评估转化为可操作的反馈
原标题：CriticGen: Generation-Aware Evaluation as Actionable Feedback
AI 摘要
CriticGen是一个细粒度、生成感知的评估框架，旨在将评估转化为可操作的答案改进控制。它首先生成样本特定的评估维度和评分标准，然后基于这些标准联合生成分数、理由、可执行的改进建议和改进后的答案。实验表明，CriticGen提高了评估相关性和覆盖率，改善了分数相关性，并将基于标准的理由和可执行建议的F1分数从0.6369/0.5994提升到0.7554/0.7900，同时改进了73.17%的答案，非退化率为93.28%。
正文节选
CriticGen: Generation-Aware Evaluation as Actionable Feedback Abstract Current evaluation methods for large language models are coarse-grained and decoupled from generation, producing generic explanations that fail to provide actionable feedback for model improvement. We propose CriticGen, a fine-grained, generation-aware evaluation framework that turns evaluation into actionable control for answer improvement. CriticGen first generates sample-specific evaluation dimensions and scoring criteria