CLASSIFICATION OF CUSTOMERS EMOTION USING NAÏVE BAYES CLASSIFIER (Case Study: Natasha Skin Care)

Abstract

Today customers can easily submit their review or opinion of a product or a service from Natasha Skin Care through mentions tweet @NatashaSkinCare. Mentions can be identified emotions of costumers after using a products or a services of Natasha Skin Care. This research proposes to classify emotions according to Ekman that is joy, surprise, anger, fear, sad, and disgust by using Naive Bayes Classifier. Naive Bayes Classifier is chosen because of its advantages that is simple, fast, and high accuracy. The dataset in this study amounted to 19,253 with the division for each class is 804 joy, 43 surprise, 154 anger, 61 fear, 287 sad, 167 disgust, and 17736 no-emotions. The results show that the Naive Bayes Classifier method has a good performance to classify the emotions Natasha Skin Care customers through twitter. The average accuracy rate on the no-emotions class dataset is 80.19%. The average of the emotional classification without involving the no-emotions class shows the highest recall value in the joy class of 92.21%. The highest precision value in the surprise class was 97.77% and the highest F1-Measure was in the joy class of 89.14%. The mean on the dataset with the no-emotions class is 88.58%. Although the accuracy of the dataset with the no-emotions class is higher, the precision and recall value is very low, which is 0%. After using the ROS resampling algorithm, the mean values of precision, recall, and F1-Measure are highest in the no-emotions class. Precision value is 96.64%, recall value is 76.36%, and F1-Measure value is 85.93%. Keyword: emotional classification, naive bayes classifier, resampling