Predicting Social Media Popularity with Multi-Task Models and Self-Attention Mechanisms

Abstract: A rising number of disciplines, including data science and marketing, are interested in predicting the popularity of social media content. We provide a novel deep learning framework in this paper that predicts many engagement metrics, including likes, comments, shares, and overall engagement, by combining self-attention processes with multi-output regression. By examining both user-specific characteristics (like follower count and account age) and post-specific data (like length, hashtags, and media presence), the model uses self-attention to dynamically allocate importance to various inputs, improving prediction accuracy and interpretability. Model training and validation are conducted using synthetic data that mimics real-world behaviour. A user-friendly online application that provides real-time forecasts and visual insights is used to install the system. The outcomes of our experiments show how well our method captures the intricate patterns that underlie social media interaction. Keywords: Social Media Prediction, Popularity Forecasting, Self-Attention Mechanism, Multi-Task Learning, Likes, Comments, Shares Prediction.