Concept driven domain adaptation improves video search for science teaching
Concept Driven Domain Adaptation: Finding an Abstract Needle in a Haystack
Computer Vision and Pattern Recognition
Summary
Science teachers often want to find short video clips illustrating scientific ideas rather than just what visually happens on screen. Existing video search tools struggle because they match words to visible events, not abstract concepts. The authors propose a new method called Concept-Driven Domain Adaptation (CDDA) that helps models connect abstract science concepts to specific video moments. This makes it easier to find relevant clips that show examples of those concepts, improving searches for educational videos.
What this means in practice
- •For educational content platforms: Enhance search tools to retrieve video clips demonstrating abstract scientific concepts by linking curriculum ideas to documentary visuals.$Commercial implications: Enables development of concept-based search engines for educational videos marketed to schools and learning platforms.
- •For video indexing teams: Improve indexing of instructional videos by bridging abstract scientific concepts with sparse, concrete video examples for better content tagging.
Tested on one dataset.
Authors
Haiming Zhao, Tai Wang, Kun Zhang, Xicheng Peng, Zhiyang Li
Abstract
Science teachers frequently search for documentary excerpts not by describing what appears on screen, but by querying the abstract concepts they intend to teach. This use case exposes a limitation of existing language-based video moment retrieval methods, which typically assume that queries describe observable events, whereas instructional search requires retrieving concrete visual phenomena that instantiate an underlying scientific principle. We study this setting as concept-to-example video retrieval, an abstract-needle-in-a-haystack problem where compact curriculum concepts must be grounded in temporally sparse documentary evidence. To bridge this abstraction gap, we propose Concept-Driven Domain Adaptation (CDDA), a three-stage framework for adapting two-tower vision-language models to concept-level retrieval. CDDA treats concepts as intermediate semantic anchors: it first structures the textual embedding space with textbook and teacher-handbook example-concept pairs, then transfers this concept-aware geometry to documentary visuals under a frozen visual encoder, and finally jointly adapts both encoders with sparse visual concept supervision. From a geometric perspective, this staged alignment reduces text-concept and vision-concept angular gaps, thereby encouraging concept-level adaptation while preserving the pretrained model's concrete image description alignment. On a curated middle-school physics retrieval benchmark, CDDA achieves stronger pedagogically oriented concept retrieval than several competitive multimodal baselines, including Qwen3-VL-Embedding-2B, while maintaining concrete image-text matching after adaptation.