Automatic Keyphrase Extraction and Segmentation of Video Lectures

Keyphrases are essential meta-data that summarize the contents of an instructional video. In this paper, we present a domain independent, statistical approach for automatic keyphrase extraction from audio transcripts of video lectures. We identify new features in audio transcripts, that capture key patterns characterizing keyphrases in lecture videos. A system for keyphrase extraction is designed that uses a supervised machine learning algorithm, based on a Naive-Bayes classifier to extract relevant keyphrases. Our extensive experimental studies show that our system extracts more relevant keywords than existing approaches. The paper also evaluates the performance of the proposed keyphrase extraction method for different categories of lectures. The extracted keyphrases are used further as features for automatic topic based segmentation of the video lectures. This process of automatic keyphrase extraction and segmentation results in a section-wise annotated video lecture which can be effectively viewed in a lecture browser.