ABSTRACT
Short video content has been proliferating like wildfire around digital platforms and with this growth comes the demand for more sophisticated predictive analytics that can be used to assess and maximize how well a given video is going to do. In this study, we present a deep learning-based end-to-end framework for predicting the overall popularity of short videos based on important engagement metrics of likes, comments, views and subscriber numbers. This study explores the three independent algorithms of deep learning i-e (CNN, Long Short-Term Memory and RNN) to cross-examine the performance and forecast capability. Each algorithm is employed to learn the complex interactions between factors of engagement and popularity, allowing a comprehensive comparative study on their predictive powers and computational costs. It consists of stepwise processing the engagement data, prominent feature extraction and then preparation of specific architectures for each algorithm as a neural network. CNN is used for hierarchical features extraction, LSTM for longterm dependencies capturing and RNN for sequential data modeling. A systematic assessment of the results to obtain the best approach for predicting video success. Our framework not only demonstrates how content creators and digital marketers can derive actionable insights from posts but also emphasizes the importance of employing different deep learning paradigms that improved prediction performance. Expected outcomes of this research will be optimizing specific content strategies, maximizing engagement potential and providing a framework for platforms to predict successful content creation.
