Aspect Based Sentiment Analysis on Shopee Application Reviews Using Support Vector Machine

Abstract

One of the e-commerce in Indonesia is Shopee. Feedback from users is needed to improve the quality of e-commerce services and user satisfaction. This research process includes data scraping, labeling, text pre-processing, TF-IDF, aspect, and sentiment classification. The novelty of this research is using the SVM method with SGD to classify Indonesian language application reviews based on aspect categories consisting of 7 dimensions of service quality and sentiment so that the website created in this research can display the aspects and sentiments of the input reviews. This research also builds an Indonesian normalization dictionary to optimize the terms used to increase model accuracy. The test in aspect classification resulted in a precision value of 90%, recall of 88.73%, accuracy of 88.57%, and f1-score of 89%. Meanwhile, the sentiment classification resulted in a precision value of 96.15%, recall of 91.91%, accuracy of 94.28%, and f1-score of 93.98%. In addition, the test results (accuracy, f1-score, precision, recall) show that the lemmatization process is better than stemming and term weighting using the TF-IDF method is better than other methods (raw-term frequency, log-frequency weighting, binary-term weighting).