Dataset Card for BloomVQA

Dataset Description

BloomVQA is a dataset based on picture stories designed for educating young children. It aims to facilitate comprehensive evaluation and characterization of vision-language models on comprehension tasks. The dataset contains tasks reflecting 6 different levels of comprehension and underlying cognitive processes, as laid out in Bloom's Taxonomy, a classic framework widely adopted in education research. This underlying hierarchical taxonomy enables graded model evaluation, automatic data augmentation and novel metrics characterizing model consistency.

The core dataset contains 1200 multiple-choice samples collected via Amazon Mechanical Turk based on 20 picture stories downloaded from Creative Commons resources Book Dash and Storyweaver.

• Paper: BloomVQA: Assessing Hierarchical Multi-modal Comprehension

Dataset Structure

Each multiple-choice sample contains 1 question and 4 free-form answers including 1 correct answer and 3 incorrect answers. Each sample is labeled with the title of picture story and the level of comprehension as defined in Bloom's Taxonomy.