CIDER

该数据集是一个经过手动整理的对话解释集，其中包含了以隐性和显性知识三元组形式推断的对话对解释，这些知识三元组是通过上下文常识推理得出的。此外，该数据集还包括按常识知识类型分类的关系（例如，因果、条件、时间）。规模上，数据集包含了807个对话，每个对话包含5到12个话语。该数据集适用于进行对话级别的自然语言推理任务、跨度提取以及多选跨度选择任务。

This dataset is a manually curated collection of dialogue explanations. It includes dialogue pair explanations inferred in the form of implicit and explicit knowledge triples derived through contextual commonsense reasoning. Additionally, the dataset covers relationships categorized by commonsense knowledge types, such as causal, conditional and temporal relations. In terms of scale, the dataset consists of 807 dialogues, with each dialogue containing 5 to 12 utterances. This dataset is applicable to dialogue-level natural language inference tasks, span extraction and multiple-choice span selection tasks.