Dataset

DrawEduMath

Hand-drawn responses to math problems annotated with QA pairs about the students' work.

52.7 MB
10 downloads
Tasks
Grading science/math
Data types
AI transcript
Image
Tabular
License
CC BY-NC-SA 4.0
Photo by Allison Shelley for the Alliance for Excellent Education

At a glance

Authored by: Worcester Polytechnic Institute and University of California Berkeley

DrawEduMath is an dataset of math problem images concatenated with students' hand-drawn responses. Each image is annotated by a human expert math teacher describing the student's written answer. Teachers additionally created QA pairs about the students' work. LLMs were called upon to make additional synthetic QA pairs. This resulted in a dataset of 2,030 images and annotations with 11,661 human-generated QA pairs and 44,362 LLM-generated QA pairs.

This dataset is useful for training models to interpret elementary students' hand-written math responses.

Permissions and licensing

License
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International
Key restrictions

This dataset is published under a non-commercial license and is available for research purposes only.

How to access the data

This dataset is available to all logged-in users.

  1. Sign in or create an account.
  2. Once you are logged in, head to the Data download tab and follow the instructions to access the data.