微软研究院多模态对齐食谱语料库

微软研究院多模态对齐食谱语料库是一个包含约15万对食谱对齐关系的数据集，涵盖4,262种不同的菜肴。该数据集通过自动对齐来自网络的文本和视频食谱，提供了丰富的常识性信息，如食材和烹饪步骤的替代描述。创建过程中，研究团队首先使用无监督算法学习不同食谱间的对齐关系，然后通过图算法实现多文本和多视频食谱的联合对齐。该数据集的应用领域包括自然语言处理、计算机视觉和机器人技术，旨在帮助机器理解和执行日常任务，如烹饪等。

The Microsoft Research Multimodal Alignment Recipe Corpus is a dataset containing approximately 150,000 pairs of recipe alignment relationships, covering 4,262 distinct dishes. By automatically aligning text and video recipes sourced from the web, this dataset provides rich commonsense information, including alternative descriptions for ingredients and cooking procedures. During its creation, the research team first used unsupervised algorithms to learn alignment relationships between different recipes, then employed graph algorithms to achieve joint alignment of multiple text and video recipes. The application fields of this dataset include natural language processing, computer vision and robotics, aiming to help machines understand and perform daily tasks such as cooking.