Evaluation of classification methods for Indonesian text emotion detection

摘要

This paper presents Indonesian text emotion detection and evaluates the performances of four different classification methods: Naive Bayes (NB), J48, K-Nearest Neighbor (KNN) and Support Vector Machine-Sequential Minimal Optimization (SVM-SMO). The experiment uses Indonesian text corpus, containing 1000 sentences which consists of six emotion classes: anger, disgust, fear, joy, sadness, and surprise. Preprocessing step which consists of tokenization, case normalization, stopword removal, stemming and TFIDF are used to extract the features of text emotion. We conduct 10-fold cross validation and split validation for the experiment. Based on the result, we conclude that SVM-SMO classifier gives the best performance. In the 10-fold cross validation, the result shows that the accuracy of NB, J48, KNN and SVM-SMO are 80.2%, 80.8%, 68.1%, and 85.5% respectively. The same conclusion is also demonstrated by the split validation, the highest accuracy of 86% is also achieved by SVM-SMO.