AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

SCRIBES: Web-Scale Script-Based Semi-Structured Data Extraction with Reinforcement Learning

Shicheng Liu, Kai Sun, Lisheng Fu, Xilun Chen, Xinyuan Zhang, Zhaojiang Lin, Rulin Shao, Yue Liu, Anuj Kumar, Wen-tau Yih, Xin Luna Dong

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
RL training with group cross-page reward vs agentic promptingQwen2.5-14B · SemiBench script gen (All)F1 (LLM-judged): 8.0 -> 19.9Improved
RL training with SCRIBES reward vs agentic baselineQwen2.5-32B · SemiBench script gen (All)F1 (LLM-judged): 19.4 -> 28.1Improved
use group-averaged reward instead of self-score only reward (ablation)Qwen2.5-14B · SemiBench HoldoutF1 (LLM-judged): 9.5 -> 16.7Improved
use group-averaged reward instead of self-score only reward (ablation)Qwen2.5-14B · SemiBench Example pagesF1 (LLM-judged): 27.9 -> 26.7Mixed
add CommonCrawl synthetic-annotation training after annotated data (ablation)Qwen2.5-32B · SemiBench (All)F1 (LLM-judged): 28.1 -> 33.2Improved
add CommonCrawl training (ablation)Qwen2.5-14B · SemiBench (All)F1 (LLM-judged): 19.9 -> 21.8Improved
restrict CC training to failure cases instead of all CC (ablation)Qwen2.5-32B · SemiBench (All)F1 (LLM-judged): 29.7 -> 33.2Improved
restrict CC training to failure cases instead of all CC (ablation)Qwen2.5-14B · SemiBench (All)F1 (LLM-judged): 22.0 vs 21.8No clear effect
train directly on CC data or 1:1 CC+annotated mix (ablation)Qwen2.5 · SemiBenchF1 (LLM-judged)No clear effect
add SCRIBES triples to QA inputGPT-4o · SemiBench QAQA accuracy (%): 82.5 -> 86.6Improved
add SCRIBES triples to QA inputQwen2.5-14B · SemiBench QAQA accuracy (%): 74.2 -> 77.3Improved
add SCRIBES triples to QA inputQwen2.5-3B · SemiBench QAQA accuracy (%): 53.8 -> 54.3No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy