Peer Review Toxic Detection Dataset
收藏资源简介:
本研究团队定义了同行评审中 toxic sentences 的四种特征,并从 OpenReview 平台收集评审,由专家根据这些定义进行标注,创建了一个用于检测同行评审中 toxic sentences 的 benchmark 数据集。该数据集包含313个句子,经过严格的两次人类标注流程以确保可靠性。数据集可用于检测模型在识别同行评审中的毒性内容的能力,旨在为构建健康、负责任的建设性学术讨论和科学合作环境做出贡献。
The research team defined four characteristics of toxic sentences in peer reviews, collected peer reviews from the OpenReview platform, and created a benchmark dataset for detecting toxic sentences in peer reviews, which was annotated by experts in accordance with these defined characteristics. This dataset contains 313 sentences and underwent a rigorous two-stage human annotation process to ensure annotation reliability. It can be used to evaluate a model's ability to identify toxic content in peer reviews, and aims to contribute to the establishment of a healthy, responsible and constructive environment for academic discussions and scientific collaboration.




