Scholarlink

<p>Lecture videos constitute the most indispensable educational resources and allow for flexible and comprehensive access to knowledge through visual and aural presentations. Topic segmentation (TS) within the videos will make segmenting content coherent within recognizable topics easier to understand and navigate. But due to domain specific problems involving terms whose meanings are complex to capture by conventional methods, topic segmentation may not be more accurate. Noisy data makes segment identification difficult. In this regard this paper proposes a model that could overcome these limitations and improve the precision as well as reliability of topic segmentation in lecture video transcripts. We propose a novel pipeline incorporating Sentence-BERT(S-BERT) embeddings, Locality Sensitive Hashing (LSH) and Hierarchical Clustering for TS. We use lecture videos available on Youtube and other platforms such as NPTEL, Khan Academy, Coursera, and Lecture VideoDB for the task, along with evaluation metrics for measuring the efficiency of topic segmentation and validating the effectiveness of the proposed approach.</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

Hierarchical clustering with SBERT and LSH to improve topic boundary segmentation accuracy in lecture videos

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • K. Vignesh,
  • S. R. Balasundaram

摘要

Lecture videos constitute the most indispensable educational resources and allow for flexible and comprehensive access to knowledge through visual and aural presentations. Topic segmentation (TS) within the videos will make segmenting content coherent within recognizable topics easier to understand and navigate. But due to domain specific problems involving terms whose meanings are complex to capture by conventional methods, topic segmentation may not be more accurate. Noisy data makes segment identification difficult. In this regard this paper proposes a model that could overcome these limitations and improve the precision as well as reliability of topic segmentation in lecture video transcripts. We propose a novel pipeline incorporating Sentence-BERT(S-BERT) embeddings, Locality Sensitive Hashing (LSH) and Hierarchical Clustering for TS. We use lecture videos available on Youtube and other platforms such as NPTEL, Khan Academy, Coursera, and Lecture VideoDB for the task, along with evaluation metrics for measuring the efficiency of topic segmentation and validating the effectiveness of the proposed approach.