An Enhanced Deep Convolutional Neural Network for YouTube Video Categorization using Textual Metadata

The rapid growth of online video platforms like YouTube has triggered an urgent need for the efficient categorization of a huge volume of videos. This paper proposes an approach to classify YouTube videos using only textual metadata like title and description using a Deep Convolutional Neural Network model. In this work, the model was trained on 20,000 videos in nine categories, where the preprocessing, tokenization, and one-hot encoding were used for better feature representation. Experimental results have shown that the DCNN outperforms RNNs, GRUs, and classic machine learning classifiers, reaching an accuracy of 96% and a ROC-AUC of 0.99. By focusing only on textual cues, the method proposed enjoys lower computing complexity compared with vision-based algorithms while retaining its predictive power.