烹饪视频中的结构化程序知识提取基准

本数据集由微软亚洲研究院创建，包含356个教学烹饪视频和15,523个视频片段/句子级标注，旨在从多模态信息中提取程序知识。数据集涵盖了89种不同的食谱类型，通过手动标注形成结构化知识，如动词-参数元组，以评估模型对程序知识的理解和提取能力。该数据集的应用领域主要集中在视频和语言理解，特别是在解决视频内容中的动作和程序理解问题，以及口头叙述中的谓词-参数结构和指代问题。

This dataset was created by Microsoft Research Asia. It contains 356 instructional cooking videos and 15,523 video clip/sentence-level annotations, aiming to extract procedural knowledge from multimodal information. The dataset covers 89 distinct recipe categories, and includes structured knowledge such as verb-argument tuples formed through manual annotation, to evaluate models' capacity for understanding and extracting procedural knowledge. Its application areas mainly focus on video and language understanding, particularly addressing action and procedural understanding tasks in video content, as well as predicate-argument structure and coreference resolution issues in verbal narratives.