arXiv cs.AI· Yufan Chen, Ching Ting Leung, Bowen Yu, Haifan Zhou, Jianwei Sun, Yong Huang, Linyan Li, Hao Chen, Hanyu Gao·· 4 小时前AI 评分58
ChemEAGLE 用多智能体系统从化学文献构建可验证数据库
Building verifiable databases at scale from the chemical literature via a multi-agent system
AI 导读
研究团队提出多智能体系统 ChemEAGLE,由工具提供化学结构、语言模型做判断,在 424 张已发表图形上实现 82.2% 的完全正确反应 F1 分数,而配备工具的通用智能体为 47.7%。其校准置信度把人工复核范围压缩到 35% 的反应,准确率达 96.6%;在无需复核的情况下,该系统构建了包含 60,103 条反应的光化学数据库,并用这些记录训练模型进行条件排序和产物预测。
正文 · 原文
Abstract:Accelerating chemical research with artificial intelligence requires high-quality databases. However, many reactions exist only as graphics drawn for human readers, which machines must read exactly. Specialist software is precise but rigid, whereas multimodal language models are flexible yet often write incorrect structures; neither flags errors. We present ChemEAGLE, a multi-agent system where tools supply structures and a language model extends their reach through judgments that programs apply. On 424 published graphics, its F1 score for exactly correct reactions was 82.2%, against 47.7% for a tool-equipped generalist agent, and calibrated confidence limited review to 35% of reactions at 96.6% accuracy. Without review, it built a 60,103-reaction photochemistry database whose records trained models to rank conditions and predict products. Verifiable databases can thus be built at scale.
| Comments: | v4: substantially revised; new title, one added author. Adds a expanded 424-graphic benchmark, generalist-agent baselines, calibrated confidence scores, and downstream tasks: condition ranking and product prediction trained on a 60,103-reaction photochemistry database. Results in v1-v3 are superseded. 92 pages, 5 figures |
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2507.20230 [cs.AI] |
| (or arXiv:2507.20230v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2507.20230 arXiv-issued DOI via DataCite |
Submission history
From: Yufan Chen [view email]
[v1]
Sun, 27 Jul 2025 11:16:57 UTC (7,650 KB)
[v2]
Tue, 29 Jul 2025 02:55:37 UTC (7,107 KB)
[v3]
Fri, 6 Mar 2026 11:15:48 UTC (12,212 KB)
[v4]
Tue, 17 Nov 2026 10:00:34 UTC (6,338 KB)
来源:arXiv cs.AI · arxiv.org