Improving Sentiment Classification and Fake Video Detection of YouTube Videos Using Custom Metadata and GAN Model

With the rapid increase in user-generated content, YouTube faces a significant challenge in managing the vast amounts of data, including the proliferation of fake videos published online in a short span of time. Effective sentiment analysis and fake video detection are crucial for understanding user engagement, preferences, and maintaining the platform's integrity. This research aims to improve sentiment classification and fake video detection for YouTube videos using custom metadata and Generative Adversarial Networks (GANs). The proposed methodology focuses on extracting sentiment from user comments and custom metadata associated with YouTube videos. A multi-stage process is employed that includes custom data creation, data preprocessing, feature extraction, and sentiment classification. By integrating domain-specific features and contextual information, our aim is to improve the accuracy and relevance of sentiment analysis. The proposed approach not only facilitates more efficient organization and categorization of YouTube content but also provides valuable insights for detecting fake videos, even with a limited number of user comments. Experimental evaluations on diverse datasets, including a custom dataset and the publicly available YouTube US video dataset, demonstrate significant enhancements in performance compared to existing methods. To further boost model performance, the GANs model is used which performs data balancing and leads to notable improvements in both recall and F1-score metrics, showing an increase of approximately 5% to 6% across all machine learning models. This research addresses the critical need for reliable sentiment analysis and fake video detection in the ever-expanding digital content landscape.