Data Labeling & Sentiment Classifier
A web app that simulates annotator work. You label Indonesian product reviews by hand, then compare your labels with an automatic classifier and see where they disagree.
What it does
- Manual labeling as positive negative neutral
- Automatic prediction from an Indonesian word lexicon, written from scratch with no NLP libraries
- Accuracy against manual labels and a list of mismatched items
- CSV export of the labeled results
How the negation rule works
tidak
bagus
bagus +1
“tidak” flips it
negative
An illustration of the rule, not a row from the dataset.
Result
Known limits
- Misses negation when the negation word is more than one word away
- No prediction when a sentence has no dictionary words
Next
Wider negation handling, a bigger lexicon, Naive Bayes or transformer models trained on labeled data, and uploading your own dataset.
The demo may take a few seconds to wake up if it has been idle.