ArtELingo-28
收藏资源简介:
ArtELingo-28是由阿卜杜拉国王科技大学创建的一个视觉语言基准数据集,涵盖了28种语言,包含约200,000个标注。该数据集基于WikiArt图像,每张图像有约140个情感标签和多语言描述,旨在捕捉不同语言和文化背景下的主观情感差异。数据集的创建过程涉及220名来自23个国家的标注者,通过Amazon Mechanical Turk平台进行标注。ArtELingo-28主要用于评估机器学习系统在多语言环境下的情感标注能力,旨在解决跨文化情感理解和表达的问题。
ArtELingo-28 is a visual-language benchmark dataset developed by King Abdullah University of Science and Technology, encompassing 28 languages and housing approximately 200,000 annotated instances. Built upon WikiArt images, each image in the dataset is associated with roughly 140 emotional labels and multilingual descriptions, with the goal of capturing subjective emotional disparities across diverse linguistic and cultural backgrounds. The dataset's annotation process recruited 220 annotators from 23 countries, who conducted the annotation tasks via the Amazon Mechanical Turk platform. ArtELingo-28 is primarily utilized to evaluate the emotional annotation performance of machine learning systems in multilingual settings, targeting the resolution of challenges related to cross-cultural emotional understanding and expression.

- 1No Culture Left Behind: ArtELingo-28, a Benchmark of WikiArt with Captions in 28 Languages阿卜杜拉国王科技大学 · 2024年



