CTBench
收藏资源简介:
CTBench是一个用于评估语言模型在临床试验设计中能力的综合基准。该数据集由伦斯勒理工学院创建,包含两个子数据集:CT-Repo和CT-Pub。CT-Repo包含1690个临床试验的基线特征,数据来源于clinicaltrials.gov;CT-Pub则是从相关出版物中人工收集的100个试验的更全面的基线特征。数据集创建过程中,使用了先进的提示工程技术来生成基线特征。CTBench的应用领域主要集中在临床试验设计,旨在通过AI辅助选择基线特征,提高临床研究的效率和鲁棒性。
CTBench is a comprehensive benchmark for evaluating the capabilities of language models in clinical trial design. Developed by Rensselaer Polytechnic Institute, this dataset comprises two sub-datasets: CT-Repo and CT-Pub. CT-Repo contains 1,690 baseline characteristics of clinical trials sourced from clinicaltrials.gov; CT-Pub, by contrast, consists of more comprehensive baseline characteristics of 100 trials manually collected from relevant academic publications. Advanced prompt engineering techniques were employed during the dataset construction process to generate the baseline characteristics. The primary application scope of CTBench focuses on clinical trial design, aiming to support AI-assisted selection of baseline characteristics and improve the efficiency and robustness of clinical research.




