Machine Learning
SENTIKEN — App Review Sentiment Analysis
Mobile app that collects and classifies the sentiment of PLN Mobile and MyPertamina Google Play reviews using TF-IDF + K-NN.
- Role
- Full-Stack & ML Developer
- Year
- 2026
- Technologies
- PythonFastAPIReact Native (Expo)TypeScriptPostgreSQLScikit-learnRedisDocker
Overview
A software implementation of a thesis methodology on app-review sentiment analysis with K-Nearest Neighbor — using no synthetic review data.
Approach
Google Play scraping with rate limiting and retries, Indonesian text preprocessing (normalisation, negation-aware stopwords, Sastrawi stemming), lexicon-based labelling, TF-IDF weighting, and K-NN with multi-K experiments.
Solution
An Expo React Native app backed by FastAPI + PostgreSQL with a Redis job queue; full evaluation (accuracy, precision, recall, F1, confusion matrix), cross-app comparison dashboard, CSV/PNG/PDF export, and JWT auth.
Model Details
- Problem
- Review sentiment classification (positive/negative, optionally neutral).
- Dataset
- Real Google Play Store reviews of PLN Mobile and MyPertamina (scraped or CSV-imported).
- Model
- TF-IDF + K-Nearest Neighbor
- Validation Strategy
- Stratified train/test split with feasibility validation.