Emotion Sentiment Analysis in Turkish Music

Music has become an indispensable element of human life. Its rhythms, melodies, and harmonies resonate deeply within us, touching our emotions and echoing our sentiments. In recent years, music emotion sentiment classifications in different languages have been studied. However, to be best of our knowledge, Turkish music has not been explored sufficiently using intelligent tools. We explore machine learning algorithms to classify Turkish music audio excerpts into distinct mood categories. We use two datasets: the Turkish Music Emotion (TME) dataset and the Turkish Emotional Voice Database (TurEV-DB) dataset. The Gradient Boosting, XGBoost, CatBoost, Random Forest, Decision Tree, and Gaussian Naive Bayes machine learning algorithms are used for training and testing the learners. Our case study results demonstrate that the CatBoost learner has the best overall performance with an Area Under the ROC Curve of 0.948 and Accuracy of about 82% for the TME dataset, and an Area Under the ROC Curve of 0.989 and Accuracy of about 90% for the TurEV-DB dataset. In the context of the two datasets, the top two learners are CatBoost followed by XGBoost. The six learners, to the best of our knowledge, have not been explored elsewhere with these two datasets, making this work a unique addition to the related literature and state-of-the-art.