Researchers said popularity did not correlate with academic value
HYDERABAD: Millions of views on a children’s video may say little about how useful it is for school learning, according to a study that examined 48,956 videos across 134 popular YouTube children’s channels and found no clear relationship between popularity and academic quality.
Researchers from the Indian School of Business (ISB) and Rochester Institute of Technology developed a machine comprehension system that tests videos against schoolwork. The model examines spoken and visual information in a video and checks whether it contains enough material to correctly answer questions drawn from textbooks. The research was published in the INFORMS Journal on Computing.
The scale of the analysis was substantial. Researchers initially collected 72,667 videos, with the final dataset comprising 48,956 videos for which both transcripts and video files were available, along with 830,481 video frames. A separate set of 1,000 videos was manually labelled for 6,355 question‑answer pairs, including 4,758 based on spoken content and 1,597 on visual information.
The videos were then tested against the ScienceQA dataset of about 21,000 textbook questions covering Classes 1 to 12, across natural science, social science and language science. It spans 379 skills, 127 categories and 26 topics.
The researchers’ best‑performing combination, which used both transcripts and visual information, achieved 77.5 per cent accuracy when three relevant videos were considered. A transcript‑based approach reached 75.21 per cent, suggesting that adding visual information improved performance, though at a higher computational cost.
The analysis also found a mismatch between what is available for younger and older learners. Units and measurement had strong coverage in Classes 1 to 5, while language areas such as literary devices and developing arguments were also well represented. More complex topics, including kinetic and potential energy, rocks and minerals, designing experiments, and phrases and clauses, had poorer coverage, particularly across higher grades.
At the channel level, Kids Learning Tube and Sesame Street were among those with higher academic quality, while ABCKidTV was cited as an example of a channel with very high viewership but low academic quality in the study’s assessment. Researchers said popularity did not correlate with academic value.
“Parents today face a digital deluge of content, making it difficult to distinguish between what is educational and what is simply engaging,” ISB Assistant Professor Sumeet Kumar said. He added that the method could help parents, educators and platforms prioritise videos that genuinely support learning.
The researchers said such academic scores could eventually become one factor in video recommendations, allowing learning value to be considered alongside engagement. The present assessment, however, is limited to English‑language textbook questions, and the authors said the method could later be extended to other languages.
