DATA–DRIVEN PREDICTION OF USER ENGAGEMENT ON FACEBOOK USING PYTHON AND ML ALGORITHMS

Abstract

The audience engagement in the modern digital media environment is an important question that allows attractively calculating its strategy in content and the efficiency of communications. This paper describes a machine learning model written in Python that attempts to predict the interaction of user with posts on Facebook which is represented in terms of number of likes, shares, and comments. Based on a well-organized data of Facebook performance indicators, we used data preprocessing and feature engineering methods and we trained and compared various predictive models. Specifically, four regression algorithms were tested including XGBoost with hyperparameter tuning, LightGBM, Random Forest, and Gradient Boosting on the potential to model the engagement behavior. The results indicate powerful predictive abilities of ensemble learning algorithms, especially predictive capabilities of analyzing patterns leading to social media reaction. This strategy proves that data science can be used in journalism and mass communication and give media professionals tools they can acting upon to better analyze their audiences, plan their editorial calendars, and execute strategic distribution of their content. The given framework can be entirely implemented in Python and transferred to other digital platforms.