BeGin

该数据集是一个以知识为基础的对话数据集，包含了来自四个对话系统在三个大型文档知识领域内的12,000条回应。这些回应被分类为完全可归因、不完全可归因和通用类别；目前只发布了开发集和测试集。该数据集的任务是检测幻觉。

This dataset is a knowledge-grounded dialogue dataset, containing 12,000 responses generated by four dialogue systems across three large-scale document-based knowledge domains. These responses are categorized into three classes: fully attributable, partially attributable, and generic. Currently, only the development set and test set have been publicly released. The task of this dataset is hallucination detection.