Music Emotion Classification Based on Indonesian Song Lyrics Using Recurrent Neural Network

Abstract

Music is one of the entertainments for the community both in Indonesia and throughout the world. Music is enjoyed in the form of instruments and has lyrics that can express emotions. The emotions produced by a song can be distinguished based on the lyrics. Songs that are in accordance with the mood are sometimes needed to enjoy music, so we need tools to distinguish emotions in the song called classification. This study uses the Recurrent Neural Network method to classify emotions based on song lyrics. The parameters of the Recurrent Neural Network that were tested in this study were hidden size, learning rate, and dropout. Data in this study were divided into development dataset and dataset testing. K-fold cross-validation is used in the model training process. The highest accuracy obtained was 82.4 percent during the testing process. Accuracy is obtained by using a hidden size parameter of 128, a learning rate of 0, 01, and a dropout of 0, 4. The highest accuracy model is used as a basis for classifying emotions based on Indonesian song lyrics into happy and sad classes. When the live process uses 10 complete songs, the average accuracy is 83.13 percent.