ME-FEVER

该数据集名为ME-FEVER，是为了进行多证据幻觉检测而设计的。它是在原始FEVER数据集的基础上合成的，旨在为模型在实际应用中提供一个更具挑战性的基准。每个实例包含两段完全无关的证据、四段部分相关的证据以及一到三段高度相关的证据。该数据集总规模为3,901个实例，其中2,663个用于训练，1,238个用于测试，任务目标是多证据幻觉检测。

This dataset, named ME-FEVER, is designed for multi-evidence hallucination detection. It is synthesized based on the original FEVER dataset, aiming to provide a more challenging benchmark for models in real-world applications. Each instance contains two completely irrelevant evidence segments, four partially relevant evidence segments, and one to three highly relevant evidence segments. The total scale of this dataset is 3,901 instances, among which 2,663 are used for training and 1,238 for testing, with the task objective being multi-evidence hallucination detection.