RUIE-Bench
收藏资源简介:
RUIE-Bench是一个由大型语言模型生成的对抗性扰动组成的新基准数据集,旨在评估通用信息提取模型的鲁棒性。该数据集由中国科学院计算技术研究所网络数据科学与技术重点实验室构建,包含14种不同扰动类型的11580个样本,覆盖了命名实体识别、关系提取和事件检测三个任务。数据集通过人工验证确保了注释的准确性,并利用大型语言模型生成多样化且真实的扰动。
RUIE-Bench is a novel benchmark dataset composed of adversarial perturbations generated by large language models, aiming to evaluate the robustness of general-purpose information extraction models. Developed by the Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences, this dataset contains 11,580 samples across 14 different perturbation types, covering three core tasks: named entity recognition, relation extraction, and event detection. The accuracy of its annotations is ensured through manual verification, and diverse and realistic adversarial perturbations are generated with large language models.

- 1Towards Robust Universal Information Extraction: Benchmark, Evaluation, and Solution中国科学院计算技术研究所网络数据科学与技术重点实验室 · 2025年



