Improving Large Language Models for Programmatic Text Understanding via Iterative Instruction RefinementLecheng YanChenyang Lyuet al.2025ICIC 2025Conference paper