AI2D-RST
收藏资源简介:
AI2D-RST数据集是由Allen Institute for Artificial Intelligence创建的多模态语料库,包含1000个英语小学科学主题的图表,如食物网、生命周期、月相和人体生理学。该数据集基于AI2D数据集,通过众包描述收集图表,旨在支持自动图表理解和视觉问答的研究。AI2D-RST数据集引入了一种新的多层注释模式,提供丰富的多模态结构描述。注释由训练有素的专家进行,描述了图表元素的组合成感知单元、由箭头和线条等图表元素建立的连接以及图表元素之间的修辞结构理论(RST)描述的论述关系。每个注释层在AI2D-RST中都使用图形表示。该语料库可供研究和教学自由使用,旨在支持图表多模态性和计算处理的实证研究。
The AI2D-RST dataset is a multimodal corpus created by the Allen Institute for Artificial Intelligence, containing 1,000 English elementary school science-themed diagrams including food webs, life cycles, lunar phases, and human physiology. Built upon the original AI2D dataset, this corpus collects diagram descriptions through crowdsourcing, aiming to support research on automatic diagram understanding and visual question answering. The AI2D-RST dataset introduces a novel multi-layer annotation schema that provides rich multimodal structural descriptions. Annotations are conducted by well-trained experts, covering the grouping of diagram elements into perceptual units, the connections established by diagrammatic elements such as arrows and lines, and the discursive relations between diagram elements described by Rhetorical Structure Theory (RST). Each annotation layer is graphically represented within the AI2D-RST corpus. This corpus is freely available for research and educational purposes, and is intended to support empirical research on diagram multimodality and their computational processing.

- 1AI2D-RST: A multimodal corpus of 1000 primary school science diagramsAllen Institute for Artificial Intelligence · 2020年



