<![CDATA[In today's digital era, sentiment analysis of product reviews on e-commerce platforms is becoming increasingly important, especially on Tokopedia, one of the largest marketplaces in Indonesia. Tokopedia provides facilities for users to leave reviews after making transactions, which play an important role in helping businesses understand customer perceptions of products. This research aims to classify the sentiment of product reviews on Tokopedia using the IndoBERT model and evaluate its performance compared to LSTM-based methods combined with FastText, Glove, and Word2Vec embedding. The LSTM-FastText model in previous research achieved the highest accuracy of 85.08%. In this study, the sentiment classification of product reviews on Tokopedia was carried out with a total of 5400 data and the sentiment classification process was divided into two categories, namely positive and negative, with the division of the dataset into three groups: training, validation, and testing. The contribution in this research is to explore the effectiveness of the IndoBERT model performance compared to previous methods that implement the LSTM model with FastText, Glove, and Word2Vec embedding. Based on the research results, the IndoBERT model achieved an accuracy of 97%, with the same F1-score value for both sentiment categories of 97%. Specifically designed with pre-training on a large Indonesian corpus, IndoBERT is able to understand the context of the text better than the LSTM model used in previous studies. This allows IndoBERT to produce higher accuracy, as it can understand product reviews in Indonesian more effectively.]]>
Copyrights © 2025
