📄 中文摘要
利用大型语言模型（LLMs）进行自动优化建模已成为支持复杂人类决策的有前景的方法。后训练已成为增强LLMs在该领域能力的关键技术，但其有效性受到高质量训练数据稀缺和未充分利用的严重制约。通过对后训练中各种问题-响应对的错误模式进行详细分析，识别出现有自动优化建模方法的两个基本限制：（L1）错误特定问题的稀疏性和（L2）与困难问题相关的稀疏奖励。这些限制可能导致性能不佳。
基于可定位错误驱动视角的自动优化建模
📄 English Summary
Automated Optimization Modeling via a Localizable Error-Driven Perspective
Automated optimization modeling using Large Language Models (LLMs) has emerged as a promising method to assist complex human decision-making. While post-training has become a crucial technique to enhance the capabilities of LLMs in this domain, its effectiveness is severely limited by the scarcity and underutilization of high-quality training data. Through a detailed profiling of error patterns across various problem-response pairs derived from post-training, two fundamental limitations of existing automated optimization modeling approaches are identified: (L1) the sparsity of error-specific problems and (L2) the sparse rewards associated with difficult problems. These limitations can lead to suboptimal performance.