Scholarlink

<p>Automatic segmentation of classroom teaching videos enables efficient access to instructional scenarios for both students and instructors. However, existing methods struggle to capture complex teacher-student interaction patterns. To address this, this paper proposes the SceSeg, a multimodal feature fusion method that integrates the text, video, audio, and speaker information to accurately segment lecture, Q&amp;A, and discussion scenarios.The method adopts a three-stage framework: multimodal feature extraction, hierarchical classification combining rule-based and probabilistic decisions, and temporal smoothing for final segmentation. Experimental results show that SceSeg achieves 71.8%, 93.3%, 81.4%, and 77.2% on ACC, WACC, MOF, and IoU, respectively, outperforming the best baseline.This work improves efficient access to classroom content and supports intelligent analysis of teaching videos.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scenario-aware unsupervised classroom video segmentation via multimodal feature fusion

  • 期刊论文

  • 发表日期：

    2026年6月26日

查看全文
  • Guoying Wang,
  • Yingjie Xia,
  • Lufeng Mo,
  • Xiaomei Yi,
  • Peng Wu

摘要

Automatic segmentation of classroom teaching videos enables efficient access to instructional scenarios for both students and instructors. However, existing methods struggle to capture complex teacher-student interaction patterns. To address this, this paper proposes the SceSeg, a multimodal feature fusion method that integrates the text, video, audio, and speaker information to accurately segment lecture, Q&A, and discussion scenarios.The method adopts a three-stage framework: multimodal feature extraction, hierarchical classification combining rule-based and probabilistic decisions, and temporal smoothing for final segmentation. Experimental results show that SceSeg achieves 71.8%, 93.3%, 81.4%, and 77.2% on ACC, WACC, MOF, and IoU, respectively, outperforming the best baseline.This work improves efficient access to classroom content and supports intelligent analysis of teaching videos.