Abstract:
Nowadays, many people change the way they listen to music by listening to the mood of the songs in the tracks. This research is interested in analyzing song extraction using natural language processing to acquire mood information. Lyrics are valuable for categorizing music. First, removing special characters and using Term-frequency/inverse-document frequency (TFIDF) and then Latent Dirichlet Allocation (LDA) are used to connect words to mood classes. We perform a lyric-based mood classification on local machine learning classifiers such as Random forest, Decision tree, Naïve Bayes, Logistic Regression, AdaBoost and XGBoost. Using grid search for tuning the best parameter yield the results XGBoost shows the highest accuracy. It can prove that boosting algorithms have better performance than local machine learning in this research. © 2021 IEEE.