The Sentiment Analysis Of Indonesian Startup Application Reviews Using TF-IDF+SVM and FastText: A Comparative Study

The rapid rise of startups in Indonesia makes user reviews on the Google Play Store a valuable data source for understanding user perceptions and satisfaction. These unstructured reviews contain insights supporting product development and business strategies. This study analyzes sentiments in Indonesian startup app reviews and compares two classification methods: TF-IDF + Linear SVM and fastText, implemented using Google Colab. Reviews were collected in September 2025 using google-play-scraper; 4,000 reviews were retrieved and refined into 3,152 unique reviews after cleaning and preprocessing. Sentiment labeling used ratings (1–2 negative, 4–5 positive); because the neutral class was limited, this study focuses on balanced binary classification with 1576 positive and 1576 negative reviews. The process involves data scraping, text preprocessing, model training, and evaluation using accuracy, precision, recall, and F1-score metrics, with Linear SVM chosen as an efficient baseline for high-dimensional sparse TF-IDF features. Results show that fastText achieves 91.88% accuracy and an F1-macro of 0.9184, slightly outperforming TF-IDF + SVM (F1-macro 0.9103), suggesting that the embedding-based approach better captures semantic nuances of Indonesian text. Future work may extend this study to ABSA to assess sentiments toward price, UI/UX, and customer service for deeper technopreneurship insights in Indonesia.