Image Popularity Prediction

Abstract

In this thesis, we compare deep learning models for the purpose of predicting the popularity of social media posts. We curated a comprehensive dataset from a renowned social media platform, encompassing a rich variety of features in- cluding images, text captions, and social attributes. Each model's performance was evaluated based on Mean Squared Error, Mean Absolute Error, and Spear- man's rank correlation coefficient. Our model, integrating convolutional neural networks for visual inputs, transformer-based models for text, and layers for social inputs, achieved a higher composite score across all evaluation metrics in contrast to the baseline model. Enhancements such as the addition of a caption network, sentiment analysis, and the removal of scaling further boosted the per- formance. This study illuminates the potential of deep learning in improving the precision of popularity prediction for social media posts. 1