sentiment-analysis-distilbert-IKN

🚩 Overview

• Base Model: cahya/distilbert-base-indonesian
• Fine-tuned on: IKN (Ibu Kota Nusantara) Dataset
• Task: Sentiment classification (Positive vs Negative)
• Language: Indonesian
• Accuracy: 91.78% on test set

---

🚀 Quickstart

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("newreyy/sentiment-analysis-distilbert-IKN")
model = AutoModelForSequenceClassification.from_pretrained("newreyy/sentiment-analysis-distilbert-IKN")

text = "IKN akan membuka banyak lapangan kerja untuk masyarakat Kalimantan."

inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
with torch.no_grad():
    logits = model(**inputs).logits
    probs = torch.softmax(logits, dim=1).squeeze()

pred = torch.argmax(probs).item()
label = "Positive" if pred == 1 else "Negative"
confidence = probs[pred].item()

print(f"Sentiment: {label} (Confidence: {confidence:.2%})")

---

🏷️ Label Description

• 0 → Negative
• 1 → Positive

---

📊 Performance Summary

---

📈 Example Classification Report

              precision    recall  f1-score   support
Negative          0.91      0.90      0.91       125
Positive          0.92      0.93      0.92       202
Overall accuracy                      0.9178

---

🗃️ Dataset Summary

• Domain: IKN-related comments from social media

• Sources: Twitter, news portals, forums

• Positive Sentiment: XX%

• Negative Sentiment: XX%

• Data Preprocessing:

  • Cleaned mentions, hashtags, emojis, and URLs
  • Lowercased text and standardized slang
  • Max input length: 512 tokens

---

⚙️ Training Configuration

📌 Limitations

• Fine-tuned exclusively on IKN-related texts, not general sentiment tasks
• Focused on Bahasa Indonesia, less optimal for mixed languages
• Classifies only positive and negative sentiments (no neutral category)

---

📜 License

Licensed under MIT — free to use for research and development.

---

📝 Citation

If you use this model, please cite:

@misc{newreyy_ikn_sentiment_distilbert,
  title={Indonesian IKN Sentiment DistilBERT},
  author={Newreyy},
  year={2024},
  howpublished={\url{https://huggingface.co/newreyy/sentiment-analysis-distilbert-IKN}}
}

And acknowledge the original DistilBERT model:

@misc{cahya_distilbert_indonesian,
  title={DistilBERT Base Indonesian},
  author={Cahya Wirawan},
  year={2021},
  howpublished={\url{https://huggingface.co/cahya/distilbert-base-indonesian}}
}

---

🤝 Acknowledgements

• Huge thanks to cahya for the base model.
• Indonesian open-source contributors and social media communities.
• Public datasets on IKN discussions used responsibly for fine-tuning.

---