DREB
收藏资源简介:
DREB是由南京大学新型软件技术国家重点实验室提出的去偏关系抽取基准数据集,旨在解决现有关系抽取模型中存在的实体偏见问题。该数据集通过实体替换技术打破实体提及与关系类型之间的伪相关性,确保模型无法仅依赖实体提及进行预测。DREB包含来自TACRED、TACREV和Re-TACRED等广泛使用的关系抽取数据集的样本,并通过Bias Evaluator和PPL Evaluator确保数据集的低偏见和高自然性。数据集的应用领域主要集中在关系抽取模型的去偏评估,旨在提升模型在真实场景中的泛化能力。
DREB is a debiased relation extraction benchmark dataset developed by the State Key Laboratory for Novel Software Technology at Nanjing University, designed to tackle the entity bias problem prevalent in contemporary relation extraction models. This dataset breaks the spurious correlation between entity mentions and relation types via entity replacement techniques, ensuring that models cannot make predictions solely by relying on entity mentions. DREB includes samples from widely used relation extraction datasets such as TACRED, TACREV, and Re-TACRED, and ensures the dataset has low bias and high naturalness through the Bias Evaluator and PPL Evaluator. The primary application scope of this dataset lies in the debiasing evaluation of relation extraction models, aiming to enhance the generalization performance of models in real-world scenarios.




