Recipe-to-Video Questions (R2VQ)

R2VQ数据集由布兰迪斯大学创建，专注于多模态自然语言处理挑战，特别是通过文本和视觉信息的实质性对齐来更好地反映动作和事件的动态。该数据集包含2000个从公开访问的食谱网站中提取的食谱，总计503,073个tokens，旨在评估NLP系统在多模态环境下的理解能力。创建过程中，数据集通过去除重复和“分布外”食谱进行预处理，并通过K-means聚类进一步筛选。R2VQ数据集的应用领域包括测试系统在理解和推理多模态信息方面的能力，特别是在理解和推理食谱中的动作和事件顺序方面。

The R2VQ dataset was created by Brandeis University, focusing on multimodal natural language processing (NLP) challenges, particularly better reflecting the dynamics of actions and events via substantial alignment between textual and visual information. This dataset contains 2000 recipes extracted from publicly accessible recipe websites, totaling 503,073 tokens, and is designed to evaluate the understanding capabilities of NLP systems in multimodal environments. During its creation, the dataset was preprocessed by removing duplicate and out-of-distribution (OOD) recipes, and further filtered via K-means clustering. The application scenarios of the R2VQ dataset include testing systems' ability to understand and reason about multimodal information, particularly regarding the sequence of actions and events within recipes.