Klasifikasi Emosi Pada Lirik Lagu Menggunakan Algoritma Multiclass SVM dengan Tuning Hyperparameter PSO

Abstract

Currently, it is increasingly difficult to determine the emotion in a song because the numbers of the songs continue to increase, based on this problem, the researcher makes a classification model using text classification. Based on these problems, this study uses the Multi Class Support Vector Machine (SVM) method with Particle Swarm Optimization (PSO) as a tuning hyperparameter and comparing the effect of 3 datasets (lines, verses, and whole songs) in the case of classifying the emotions of song lyrics. In this case, there are five basic human emotions, in-between love, happiness, anger, fear, and sad. Based on the test results on each model, scenario 2 (SVM-PSO Perbaris) does provide the best model performance with an accuracy value of 92.13%. However, if we look at the performance value, it changes from the evaluation of the training data to the testing data presented in table 4.3, the most significant changes occur in the verses dataset and the whole song dataset. This can happen because the content or value of the per-bait dataset and the whole song has more sentences than the per-line dataset. So that the quality will be better if you use the verses dataset or the whole song. This research has also succeeded in make the classification of emotions so that it can classify the class of emotions from the text of Indonesian song lyrics.