Muhammad Fajri Davyza Chaniago, Shintami Chusnul Hidayati, Zulfayanti Sofia Solichin, Abdul Munif, Yeni Anistyasari
Predicting social media engagement has become increasingly important for content creators, marketers, and consumer technology developers seeking to enhance user experience and optimize content strategies. While existing studies primarily focus on user-generated content (UGC), non-user-generated content (Non-UGC) features, such as user attributes and metadata, remain underexplored. This study investigates the impact of Non-UGC features - including 'Following", 'Followers", 'Sponsored Label', 'Influencer Category', and 'Timestamp' - on the performance of various machine learning models. We evaluate five classifiers: Decision Tree, Random Forest, CatBoost, Support Vector Regressor, and Artificial Neural Network, chosen for their range from simple, interpretable models to complex, non-linear algorithms. The results indicate that each Non-UGC feature contributes differently to model performance, highlighting the varying importance of these features in predicting social media engagement. © 2024 IEEE.
Institut Teknologi Sepuluh Nopember, Department of Informatics, Surabaya, Indonesia; Universitas Negeri Surabaya, Department of Informatics, Surabaya, Indonesia
