ai-digest.dev
last updated 13 h ago
MultimodalarXiv cs.AI 7 d ago

GeoDial: A Multimodal Conversational Tutoring Dataset for Geometry Problem-Solving with Visual Tutor Turns

GeoDial is a new multimodal dataset comprising over 1,300 teacher-student dialogues specifically for geometry problem-solving, incorporating visual elements through diagram highlights. It features a novel annotation protocol that combines dialog acts with visual feedback, allowing for detailed supervision of both language and visual tutoring. Initial experiments with fine-tuning vision-language models on GeoDial demonstrate improvements in dialog generation but reveal challenges in accurately producing diagram highlights, underscoring the necessity for enhanced methods that integrate visual reasoning within educational AI applications.

geometrydatasetvisual tutoringrelevance 0.00 · engagement 0.00
Read at source ↗← all news
GeoDial: A Multimodal Conversational Tutoring Dataset for Geometry Problem-Solving with Visual Tutor Turns — AI News Digest