Indonesian sentiment analysis with IndoBERT

README.md

Indonesian Sentiment Analysis with IndoBERT

A small Python example for classifying Indonesian text sentiment with a Hugging Face Transformers pipeline.

The script uses mdhugol/indonesia-bert-sentiment-classification, a sentiment classification model fine-tuned from IndoBERT.

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run

python app.py

The first run downloads the model from Hugging Face.

Labels

The model returns generic classifier labels, so app.py maps them to readable sentiment labels:

Model label	Sentiment
LABEL_0	positive
LABEL_1	neutral
LABEL_2	negative

Use in Code

from app import classify_sentiment

result = classify_sentiment("Pelayanannya sangat baik dan prosesnya cepat.")
print(result)

Example output:

{
    "text": "Pelayanannya sangat baik dan prosesnya cepat.",
    "label": "positive",
    "score": 0.99,
}

app.py

from typing import Any, Dict

from transformers import pipeline

MODEL_NAME = "mdhugol/indonesia-bert-sentiment-classification"

LABELS = {
    "LABEL_0": "positive",
    "LABEL_1": "neutral",
    "LABEL_2": "negative",
}


classifier = pipeline("sentiment-analysis", model=MODEL_NAME, tokenizer=MODEL_NAME)

EXAMPLES = [
    # Formal
    "Pelayanannya sangat baik dan prosesnya cepat.",
    "Aplikasi ini sering error dan sangat mengecewakan.",
    "Menurut saya produknya biasa saja, tidak terlalu bagus tapi juga tidak buruk.",
    # Informal
    "barangnya mantappp, pengiriman cepet, seller ramah bgt 👍",
    "parah sih ini, udah nunggu lama eh barangnya rusak.",
    "lumayan lah, ga jelek-jelek amat.",
    "gw kira bakal bagus, ternyata zonk wkwk.",
    "adminnya slow response banget, bikin males.",
    "worth it sih dengan harga segini.",
    "nggak nyesel beli di sini, recommended!",
    "fiturnya oke, tapi UI-nya agak ribet.",
    "yaudah lah, standar aja menurutku.",
    "anjir bagus banget hasilnya, di luar ekspektasi.",
    "ga sesuai foto, kecewa berat.",
    "kirain abal-abal, ternyata legit juga.",
    "bagus sih, cuma packaging-nya kurang aman.",
    "not bad, tapi masih banyak bug.",
]


def classify_sentiment(text: str) -> Dict[str, Any]:
    result = classifier(text)[0]
    return {
        "text": text,
        "label": LABELS.get(result["label"], result["label"]),
        "score": result["score"],
    }


if __name__ == "__main__":
    for text in EXAMPLES:
        result = classify_sentiment(text)
        print(f"Text: {result['text']}")
        print(f"Sentiment: {result['label']} ({result['score']:.2%})")
        print()

requirements.txt

transformers>=4.40,<5
torch>=2.2
sentencepiece>=0.2