Mood classification from song lyrics using the Naive Bayes Algorithm, Support Vector Machine (SVM) and XGBoost

Abstract

Music has become a crucial aspect of daily life, with more people using streaming apps to listen to music. This research focuses on analyzing song extraction through Natural Language Processing (NLP) to determine mood information. The primary feature used in this analysis is song lyrics. Initially, we perform text preprocessing and apply feature extraction methods like Term Frequency-Inverse Document Frequency (TF-IDF), Smoothed TF-IDF, Inverse Term Count (ITC), and Smoothed ITC. Subsequently, we classify moods based on lyrics using machine learning classifiers, including Naïve Bayes, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) algorithms. The findings of this study reveal that the SVM algorithm excels over other feature extraction methods. The most accurate combination is Smoothed ITC with the SVM algorithm, achieving an accuracy of 91.52%, precision of 91.84%, recall of 91.40%, and F1-score of 91.50%.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint