CHEMRAG-BENCH
收藏资源简介:
CHEMRAG-BENCH是一个全面的基准测试,旨在系统地评估检索增强生成(RAG)方法在化学领域的有效性。该基准包含1,932个专家审核的问题-答案对,涵盖了从描述引导分子设计、逆合成、化学计算到分子注释、命名转换和反应预测等多种化学相关任务。数据集整合了科学文献、PubChem数据库、PubMed摘要、教科书和维基百科条目等多种异构知识来源。CHEMRAG-BENCH旨在为化学领域的RAG系统提供系统性的评估标准,并指导未来的研究。
CHEMRAG-BENCH is a comprehensive benchmark designed to systematically evaluate the effectiveness of retrieval-augmented generation (RAG) methods in the field of chemistry. It contains 1,932 expert-validated question-answer pairs, covering a wide range of chemistry-related tasks spanning from description-guided molecular design, retrosynthesis and chemical calculations to molecular annotation, nomenclature conversion and reaction prediction. The dataset integrates diverse heterogeneous knowledge sources including scientific literature, PubChem database, PubMed abstracts, textbooks and Wikipedia entries. CHEMRAG-BENCH aims to provide a systematic evaluation standard for RAG systems in the chemistry domain and guide future research.

- 1Benchmarking Retrieval-Augmented Generation for Chemistry伊利诺伊大学香槟分校计算机与数据科学学院, 美国国立卫生研究院国家医学图书馆 · 2025年



