Seamless Lecture Navigation Starts Here: The EDU-V Benchmark for Concept-Centric Video Understanding
IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 17525-17542
DOI: 10.1109/tcsvt.2026.3721402 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Jie Zhang und 7 weiteren Autor:innen. Die Arbeit nennt 73 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Excellent Young Scientists Fund of Jiangsu Natural Science Foundation.
Abstract (Originalsprache)
Navigating lengthy lecture videos to find specific knowledge concepts remains a barrier for learners, forcing them to manually scroll through progress bars to locate the video segments that explain specific knowledge concepts. To enable concept-centric video understanding—automatically associating videos with their underlying knowledge concept—we introduce EDU-V, a large-scale multimodal benchmark, featuring 10,000+ concept-annotated segments from online educational videos. Our dataset benchmark is built through a Large Language Model-empowered annotation pipeline for high-quality labels with minimal manual intervention. Comprehensive validation shows that EDU-V achieves high quality, evidenced by a high visual score of 4.74/5 and good textual fluency with a low perplexity of 15.54. To explore the utility of EDU-V, we propose a multimodal method named CORE (COmplementary REpresentation via parameter interaction) for knowledge concept localization in educational videos. Extensive experiments on EDU-V show that CORE outperforms the compared baselines, achieving a 19% improvement in Top-1 Recall.
Förderung
- National Natural Science Foundation of China
- Excellent Young Scientists Fund of Jiangsu Natural Science Foundation
Zitieren
APA
Zhang, J., Wang, W., Yang, W., Mao, R., Yuan, Y., Zhang, B., Gao, Y., & Yang, M. (2026). Seamless Lecture Navigation Starts Here: The EDU-V Benchmark for Concept-Centric Video Understanding. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 17525-17542. https://doi.org/10.1109/tcsvt.2026.3721402
Vancouver
Zhang J, Wang W, Yang W, Mao R, Yuan Y, Zhang B, et al. Seamless Lecture Navigation Starts Here: The EDU-V Benchmark for Concept-Centric Video Understanding. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):17525-17542. doi:10.1109/tcsvt.2026.3721402
BibTeX
@article{Zhang2026,
author = {Zhang, Jie and Wang, Wei and Yang, Wanqi and Mao, Rui and Yuan, Yi and Zhang, Bo and Gao, Yang and Yang, Ming},
title = {Seamless Lecture Navigation Starts Here: The EDU-V Benchmark for Concept-Centric Video Understanding},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
year = {2026},
volume = {36},
number = {11},
pages = {17525--17542},
doi = {10.1109/tcsvt.2026.3721402}
}
EUVIMED