DuExplain阅读理解可解释评测数据集

DuExplain阅读理解可解释评测数据集旨在评估模型预测依赖证据的可解释性，提供了人工标注的细粒度证据和扰动数据，利用证据匹配F1 Score、扰动下证据一致性、证据充分性等指标评估证据的合理性和忠诚性；同时，该数据集提供了中英文标注数据。

The DuExplain Reading Comprehension Explainability Evaluation Dataset is designed to evaluate the explainability of model predictions that rely on evidence. It provides manually annotated fine-grained evidence and perturbation data, and uses metrics including Evidence Matching F1 Score, evidence consistency under perturbations, and evidence sufficiency to assess the rationality and faithfulness of the evidence. Meanwhile, this dataset offers annotated data in both Chinese and English.