<![CDATA[User reviews on Google Play Store provide valuable information for understanding user perception of mobile applications, including children's educational games. However, review data are commonly unstructured, informal, and difficult to analyze manually. This study aims to classify the sentiment of Baby Panda School Bus game reviews using lexicon-based pseudo-labeling, word n-gram TF-IDF, and XGBoost. The dataset was collected from Google Play Store and contained 6,299 initial reviews. After duplicate removal, text cleaning, normalization, tokenization, stopword removal, and stemming, 5,337 reviews were used for modeling. Sentiment labels were generated automatically using the InSet lexicon into positive, neutral, and negative classes. The dataset was split into 80% training data and 20% testing data using stratified sampling. The selected text representation was word n-gram TF-IDF with unigram, bigram, and trigram features, followed by XGBoost classification. The experimental results achieved an accuracy of 85.86%, macro F1-score of 83.57%, and weighted F1-score of 85.61%. These findings indicate that n-gram TF-IDF and XGBoost can effectively classify lexicon-based sentiment patterns in Indonesian mobile application reviews.]]>
Copyrights © 2026
