Multiclassifier for Aspect-Based Sentiment Analysis on Indonesian Reviews of Kredit Pintar Online Lending App

Aspect-based sentiment analysis is carried out to determine user sentiment's polarity while still considering the aspects of the discussion so that all information in the reviews can be represented. The multiclassifier in this research means that two classifiers are used, the first is an aspect classifier to perform aspect classification, and the second is a sentiment classifier to perform sentiment classification based on its aspect. Two approaches were studied according to their classification work, namely, individual model in the form of an aspect classification and sentiment classification based on the aspects individually and sequentially model by testing the aspect and sentiment models sequentially. This study evaluates the performance of Support Vector Machine (SVM) and Naïve Bayes Classifier (NBC) algorithms and their combination in sequential testing. In addition, to overcome imbalance dataset, the use of oversampling types is also compared. In the individual approach, the SVM model with random oversampling showed the best performance with average accuracy, precision, recall, and f1 score of 92%, 86%, 80%, and 83% for the aspect model and 86%, 81%, 80%, and 80% for the sentiment model, respectively. While in the sequential approach, the combination of SVM-SVM with oversampling shows the best performance with an average accuracy, precision, recall, and f1-score of 79%, 56%, 50%, and 53%, while the performance of other model combinations is only in the maximum range of 77% accuracy, 55% precision, 51% recall, and 52% f1-score.