@inproceedings{ding-etal-2023-ji,
    title = "基于推理链的多跳问答对抗攻击和对抗增强训练方法(Reasoning Chain Based Adversarial Attack and Adversarial Augmentation Training for Multi-hop Question Answering)",
    author = "Ding, Jiayu  and
      Wang, Siyuan  and
      Wei, Zhongyu  and
      Chen, Qin  and
      Huang, Xuanjing",
    editor = "Sun, Maosong  and
      Qin, Bing  and
      Qiu, Xipeng  and
      Jiang, Jing  and
      Han, Xianpei",
    booktitle = "Proceedings of the 22nd Chinese National Conference on Computational Linguistics",
    month = aug,
    year = "2023",
    address = "Harbin, China",
    publisher = "Chinese Information Processing Society of China",
    url = "https://aclanthology.org/2023.ccl-1.1",
    pages = "1--16",
    abstract = "{``}本文提出了一种基于多跳推理链的对抗攻击方法,通过向输入文本中加入对抗性的攻击文本,并测试问答模型在干扰数据下生成答案的准确性,以检测问答模型真正执行多跳推理的能力和可解释性。该方法首先从输入文本中抽取从问题实体到答案实体的推理链,并基于推理链的特征把多跳问题分为了不同的推理类型,提出了一个模型来自动化实现问题拆解和推理类型预测,然后根据推理类型对原问题进行修改来构造攻击干扰句。实验对多个多跳问答模型进行了对抗攻击测试,所有模型的性能都显著下降,验证了该攻击方法的有效性以及目前问答模型存在的不足;向原训练集中加入对抗样本进行增强训练后,模型性能均有所回升,证明了本对抗增强训练方法可以提升模型的鲁棒性。{''}",
    language = "Chinese",
}