Word-level Adversarial Example Detection Dataset

本数据集由首尔国立大学创建，旨在促进文本分类中单词级对抗样本检测的研究。数据集涵盖了四种流行的攻击方法在四个不同的文本分类数据集和四个模型上的应用，总计30种组合。创建过程涉及对原始文本进行对抗性修改，以生成对抗样本。该数据集的应用领域包括自动化任务，如评论情感分析，旨在通过检测对抗性输入来增强模型的鲁棒性。

This dataset was developed by Seoul National University to advance research on word-level adversarial example detection for text classification. It includes a total of 30 unique combinations, which are created by applying four popular adversarial attack methods across four distinct text classification datasets and four different models. The dataset construction process involves generating adversarial examples through adversarial modifications to the original text. Application scenarios of this dataset cover automated tasks such as comment sentiment analysis, with the core goal of enhancing model robustness by detecting adversarial inputs.