<![CDATA[The growth of e-commerce has generated many user reviews, which are an important source for understanding consumer satisfaction and perceptions. However, manual analysis of unstructured reviews that use informal language is ineffective. In addition, conventional sentiment analysis approaches are often unable to capture the linguistic variations of the Indonesian language. This study uses the IndoBERT contextual language model to classify the sentiment of e-commerce application reviews on Shopee, Tokopedia, and Lazada. Data was collected through web scraping, amounting to 12,000 data points, with 4,000 for each application, labeled based on ratings, processed through preprocessing stages, balanced using Random Oversampling, and trained for three-class sentiment classification. The evaluation showed an Macro F1-Score of 0.90, indicating strong performance across all sentiment classes, including minority classes. These results confirm the effectiveness of IndoBERT in handling data imbalance in Indonesian sentiment analysis.]]>
Copyrights © 2025
