CONSPIRED (CONSPIR Evaluation Dataset)
收藏资源简介:
CONSPIRED 数据集是一个关于阴谋论文章的认知特征的标注数据集,用于捕捉在线阴谋论文章中多句话摘录 (80-120 个单词) 的认知特征,并使用 CONSPIR 认知框架 (Lewandowsky 和 Cook, 2020) 进行标注。CONSPIRED 是第一个标注了普遍认知特征的阴谋论内容数据集。该数据集旨在支持对阴谋论推理模式的计算分析,并评估大型语言模型对阴谋论输入的鲁棒性。
The CONSPIRED dataset is an annotated dataset targeting the cognitive characteristics of conspiracy theory articles. It is designed to capture the cognitive features of multi-sentence excerpts (80–120 words) extracted from online conspiracy theory articles, and annotated via the CONSPIR cognitive framework (Lewandowsky and Cook, 2020). As the first conspiracy theory content dataset annotated with universal cognitive characteristics, CONSPIRED aims to support computational analyses of conspiracy theory reasoning patterns and evaluate the robustness of large language models when processing conspiracy theory inputs.

- 1ConspirED: A Dataset for Cognitive Traits of Conspiracy Theories and Large Language Model Safety德国达姆施塔特工业大学 Ubiquitous 知识处理实验室 (UKP Lab), 计算机科学与黑森州人工智能中心 (hessian.AI), 德国应用网络安全国家研究中心 ATHENE, 阿联酋 Mohamed bin Zayed 人工智能大学 · 2025年



