AdvRACE

AdvRACE是一个用于评估机器阅读理解模型鲁棒性的新型基准数据集，由哈尔滨工业大学社会计算与信息检索研究中心创建。该数据集包含4934个样本，源自RACE数据集，通过引入四种不同类型的对抗攻击进行增强。AdvRACE旨在通过模拟真实世界中的文本扰动，评估模型在面对复杂和多变的输入时的表现。数据集的创建过程涉及对原始文本的多种修改，如添加干扰信息、字符交换等，以生成具有挑战性的测试案例。AdvRACE的应用领域主要集中在提高机器阅读理解模型的鲁棒性，解决模型在实际应用中可能遇到的输入扰动问题。

AdvRACE is a novel benchmark dataset for evaluating the robustness of machine reading comprehension models, developed by the Social Computing and Information Retrieval Research Center of Harbin Institute of Technology. This dataset comprises 4,934 samples derived from the RACE dataset, which is augmented with four distinct types of adversarial attacks. AdvRACE aims to assess model performance when confronted with complex and variable inputs by simulating real-world text perturbations. The dataset creation process entails multiple modifications to the original text, such as adding distracting information and character swaps, to generate challenging test cases. The core application scenarios of AdvRACE focus on enhancing the robustness of machine reading comprehension models and addressing input perturbation issues that models may encounter in practical applications.