A machine learning model to detect early math content in YouTube videos
Authors:
- Claire ChristensenClaire Christensen
- Madeline CincebeauxMCMadeline Cincebeaux
- Anirban RoyARAnirban Roy
- Sujeong KimSKSujeong Kim
Open Access
- 15 September 2025
- journal article
- Published by Emerald Publishing in Artificial Intelligence in Education
- Vol. 1 (1) , 75-92
Abstract
Purpose: Supporting the development of math skills in early childhood has wide-ranging benefits. Exposure to high-quality screen media is a convenient and effective way to promote early math skills. Yet as children’s video consumption has shifted away from television and toward online streaming platforms, it is more difficult to understand the extent to which they engage with high-quality early math media. We developed the first machine learning tool to automatically detect prekindergarten and kindergarten math content in online videos for young children. Design/methodology/approach: We describe our approach to developing a multimodal machine learning model. The process included codebook creation, video annotation, model development, and model training and testing. To test the model, we compared its performance for classifying video content to that of human annotators, in both a curated and a naturalistic dataset. Findings: Compared with human annotators, the model detects early math content with 92% precision in a curated dataset and with 63% precision when adapting to a new dataset of videos watched by children. We describe the model’s precision for detecting math content categories and indicators of pedagogical quality. Further, 79% of videos in the curated dataset that included at least one early math content category also included at least one pedagogical quality indicator. Originality/value: This is the first machine learning model to detect standards-aligned early math content in videos. Because this method is more efficient than human content classification (the current standard practice), it can better keep pace with the rate of online video production. In the future, developers could use this capability to create a content recommender tool for parents and educators. Developers could also create tools for researchers to better understand the effects of online video exposure on children’s early math learning. These findings offer a scalable new method for understanding and improving young children’s exposure to high-quality educational media.Keywords
Funding Information
- National Science Foundation (2139219)
This publication has 23 references indexed in Scilit:
- Evaluating YouTube videos for young childrenEducation and Information Technologies, 2020
- A curriculum supplement that integrates transmedia to promote early math learning: A randomized controlled trial of a PBS KIDS interventionEarly Childhood Research Quarterly, 2019
- Measuring with Murray: Touchscreen technology and preschoolers' STEM learningComputers in Human Behavior, 2016
- An overview of learning analyticsTeaching in Higher Education, 2013
- Effects of Sesame Street: A meta-analysis of children's learning in 15 countriesJournal of Applied Developmental Psychology, 2013
- The impacts of an early mathematics curriculum on oral language and literacyEarly Childhood Research Quarterly, 2012
- Early math matters: Kindergarten number competence and later mathematics outcomes.Developmental Psychology, 2009
- School readiness and later achievement.Developmental Psychology, 2007
- Developmental implications of commercial broadcasters' educational offeringsJournal of Applied Developmental Psychology, 2001
- Diversity of decision-making models and the measurement of interrater agreement.Psychological Bulletin, 1987