Klasifikasi Nuansa Emosi Film Berdasarkan Sinopsis Menggunakan Logistic Regression

Abstract

With the rapid development of digital era, movie industry has consider to be a dominant form of entertainment. However, with thousands of movies released every year, audiences often face difficulties selecting movies based on their emotional preferences. This study proposes a classification approach to determine the emotional nuance of movies (happy, sad, tense) based on their synopsis. The dataset used in this study was sourced from Kaggle with 676.491 entries and labeled using a DistilBERT pre-trained emotion detection model from Hugging Face. After mapping the labeled entries and undersampling, 6.600 balanced samples were used. Following preprocessing and data splitting, TF-IDF text representation and Logistic Regression model were applied. The model achieved 76% accuracy on validation data and improved to 83% on test data, with macro F1-scores reflecting consistent performance across all classes. These results suggest that movie synopses contain sufficient emotional signals to be automatically classified using a lightweight and effective machine learning approach.