Benchmark tests how well AI understands teacher student video interactions
VIRA-Bench: Diagnosing Pedagogical Contingency Understanding in Online Tutoring Videos
Human-Computer Interaction
Summary
Watching a tutoring video means understanding not just what happens, but how a teacher’s actions lead to student responses and then how the teacher adjusts afterward. The authors created VIRA-Bench, a test dataset of short tutoring clips paired with questions that check if AI models can figure out these important teaching moments. They found AI models still struggle at this, performing much worse than humans, especially at reasoning about how teacher and student actions depend on each other over time. This benchmark helps show where current models fail and guides better development of AI that understands tutoring videos deeply.
What this means in practice
- •For educational technology developers: Test AI models' ability to understand teacher-student interactions in tutoring videos for better automated tutoring feedback systems.
- •For multimodal ai engineers: Use VIRA-Bench to improve multimodal dialogue systems by targeting the specific challenge of linking actions and responses over short video clips.
Authors
Weihang Wang, Hansu Gu, Guanming Liu, Peng Zhang, Ning Gu, Tun Lu
Abstract
Understanding a one-to-one tutoring video requires more than recognizing its events. A model must connect a visible teacher intervention to the learner response it elicits and to the follow-up adjusted to that response. Existing video and educational-dialogue benchmarks typically emphasize event recognition, temporal reasoning, social-state inference, or transcript-based interaction, but rarely diagnose this action-response-adjustment relation. We introduce VIRA-Bench, a focused diagnostic benchmark for short-horizon pedagogical contingency in one-to-one online tutoring. VIRA-Bench contains 1,005 tutoring clips and 1,322 multiple-choice questions with researcher evidence audit and expert verification, organized into Observable Evidence Grounding (OEG), Interaction Contingency Reasoning (ICR), and Support-Response Modeling (SRM). Across 11 models, the best reaches 61.3%, well below the 91.0% human reference on a stratified subset, and ICR generally has the lowest layer accuracy among leading models. Under controlled input manipulations, removing video most strongly degrades OEG, whereas disrupting temporal order most strongly degrades ICR. Free-response, Stem+Options, and synchronized-shuffle controls show that response format, static question cues, and video-ASR mismatch do not explain these layer-specific patterns. The results identify distinct behavioral evidence dependencies without claiming an internal architectural mechanism. VIRA-Bench provides a focused evaluation resource for multimodal dialogue grounding, pedagogical discourse, and temporal reasoning in online tutoring.