Researchers have developed a new method for extracting semi-structured data from web pages. SCRIBES uses reinforcement learning to generate reusable scripts that can be applied to groups of similar web pages, improving extraction quality and efficiency. The approach outperforms existing methods by over 13% in script quality and boosts downstream question answering accuracy by more than 4%. The method is trained on synthetic annotations from CommonCrawl data. --- Why it matters: This matters because it enables scalable and resource-efficient web information extraction, which is crucial for applications like question answering and data integration. It also shows the potential of reinforcement learning in addressing complex web data extraction tasks. Source: https://arxiv.org/abs/2510.01832