YouTube Video Categorization Based on Closed Captions

YouTube, the world's most popular video-sharing platform, has over 2.6 billion unique monthly viewers and uploads 500 hours of new content every minute. Effective classification is critical for improving discoverability and viewer satisfaction among various videos. Previous research has investigated numerous methods, including keywords, frames, titles, descriptions, speech recognition, and comments, using algorithms such as Latent Dirichlet Allocation (LDA), Random Forest Classifier, and Convolutional Neural Networks. This study presents an innovative and comprehensive methodology for categorizing English YouTube videos using Closed Captions (CC) given by YouTube and LDA. A dataset was methodically built by gathering YouTube videos with CC with a maximum of 800 words. The data was recorded in JSON format, preprocessed, and then categorized using LDA. The algorithm found ten distinct topics, each representing a cohesive cluster of interconnected notions. Closed Captions provides a distinct analytical perspective, allowing deep and diversified insights into the videos' content. This novel approach contributes to the area by proposing a robust method for video categorization that takes advantage of language signals within Closed Captions. This strategy improves classification accuracy and allows for a more nuanced comprehension of the underlying themes in video content. The ability to distinguish ten unique subjects improves the categorization process, resulting in a more detailed and informative classification of videos on the platform. The findings promise to provide content creators, consumers, and researchers with valuable insights into YouTube's complex tapestry of content.