CLASSIFICATION OF SONG MOOD BASED ON LYRIC ANALYSIS USING TF-IDF AND SUPPORT VECTOR MACHINE

Abstract

The mood of a song affects listeners' perception and experience of music, but manual mood identification has become less efficient with the increasing availability of digital songs on streaming platforms. This study aims to automatically classify song moods based on lyric analysis using the Term Frequency–Inverse Document Frequency (TF-IDF) method and the Support Vector Machine (SVM) algorithm. The dataset used consists of a collection of song lyrics that have been processed through text preprocessing stages, including text cleaning, letter normalization, and removal of irrelevant common words. Lyric features are extracted using TF-IDF and used as input in the SVM model training process to predict the mood of unlabeled songs. The classification results show that moods with high emotional intensity, such as angry, excited, and happy, dominate the predictions, while romantic and sad moods are fewer in number. Model performance evaluation using accuracy, precision, recall, and F1-score metrics shows that SVM is capable of classifying song moods based on lyrics with stable and reliable performance. These findings indicate that the combination of TF-IDF and SVM is effective for automatic song mood identification and has the potential to be further developed in emotion-based music recommendation systems, thereby facilitating digital platforms in presenting music content that suits listeners' emotional preferences.