📰 Burmese News Classification with BiLSTM & Streamlit

This project builds an AI-powered Burmese news classifier using both traditional machine learning (Naïve Bayes) and deep learning (BiLSTM).
It includes a ready-to-run Streamlit web app for real-time predictions in the Burmese language.

---

🧩 Project Overview

• Classifies Burmese news articles into categories such as Politics, Business, Sports, and Entertainment.
• Implements custom Burmese NLP preprocessing — tokenization, stopword removal, and sequence padding.
• Compares performance between Naïve Bayes (TF-IDF) and BiLSTM (Keras).
• Provides an interactive Streamlit UI for user testing and visualization.

---

📊 Model Highlights

Model	Type	Accuracy	Description
Naïve Bayes	ML baseline	~80%	TF-IDF with sklearn
BiLSTM	Deep Learning	~90%	Sequence model using TensorFlow/Keras

---

📁 Project Structure

├── app.py                        # Streamlit web app
├── Burmese_News_Classification.ipynb  # Training notebook
├── requirements.txt              # Dependencies
├── stopwords.txt                 # Burmese stopword list
├── models/                       # Trained models
│   ├── bilstm_mynews.keras
│   ├── nb_tfidf.joblib
│   ├── keras_tokenizer.pkl
│   ├── label_encoder.pkl
│   └── config.json
└── README_streamlit.md           # Detailed Streamlit setup

---

🚀 Running the Streamlit App

pip install -r requirements.txt
streamlit run app.py

Then open the provided local URL (e.g. http://localhost:8501) in your browser.

---

🧠 Example Predictions

Input Text	Predicted Category
"အဆိုတော်အသစ်တစ်ဦး ပြိုင်ပွဲတွင် ပထမဆုရရှိခဲ့သည်။"	Entertainment
"အစိုးရသည် စီးပွားရေးဖွံ့ဖြိုးရေးအတွက် ငွေထုတ်ပေးမည်။"	Business

---

🧰 Technologies Used

• Python 3.10
• TensorFlow / Keras
• Scikit-learn
• Pyidaungsu Tokenizer
• Streamlit for deployment

---

🔮 Future Enhancements

• Integrate Multilingual BERT (mBERT) for improved contextual understanding.
• Add Gradio/Flask API endpoint for backend usage.
• Extend to multi-label classification (e.g., News + Sentiment).

---