amine77776/imdb-sentiment-nn

IMDB Sentiment Analysis — Neural Network Classifier

Sentiment analysis of IMDB movie reviews using a TF-IDF + Feedforward Neural Network trained automatically via GitHub Actions CI/CD and deployed to Hugging Face Hub.

---

🏗️ Architecture

IMDB Review Text
      ↓
Text Cleaning (HTML tags, punctuation, lowercase)
      ↓
TF-IDF Vectorization (10,000 features, unigrams + bigrams)
      ↓
Linear(10000 → 512) + BatchNorm + ReLU + Dropout(0.3)
      ↓
Linear(512 → 128) + BatchNorm + ReLU + Dropout(0.3)
      ↓
Linear(128 → 2) → Softmax
      ↓
Positive 😊 / Negative 😞

---

📊 Dataset

• File: imdb_balanced_10k.csv
• Size: 10,000 reviews (5,000 positive, 5,000 negative)
• Source: IMDB movie reviews

---

🚀 CI/CD Pipeline

Every push to main automatically:

1. ✅ Installs Python dependencies
2. ✅ Trains the neural network on the full dataset
3. ✅ Saves model artifacts (model.pt, vectorizer.pkl, config.json, metrics.json)
4. ✅ Uploads everything to Hugging Face Hub

No manual uploads. Ever.

---

📁 Project Structure

imdb-sentiment-nn/
├── data/
│   └── imdb_balanced_10k.csv   # Training data
├── model/                       # Generated by train.py
│   ├── model.pt                 # PyTorch model weights
│   ├── vectorizer.pkl           # TF-IDF vectorizer
│   ├── config.json              # Hyperparameters
│   └── metrics.json             # Training metrics
├── train.py                     # Training script
├── predict.py                   # Inference script
├── requirements.txt             # Dependencies
└── .github/
    └── workflows/
        └── train-and-upload.yml # CI/CD pipeline

---

🔧 Local Usage

Install dependencies

pip install -r requirements.txt

Train the model

python train.py

Run inference

python predict.py "This movie was absolutely fantastic!"
# Sentiment:   positive 😊
# Confidence:  94.23%

python predict.py "Terrible film, complete waste of time."
# Sentiment:   negative 😞
# Confidence:  91.87%

---

📈 Results

See model/metrics.json after training. Expected accuracy: ~88–92% on the test set.

---

🤗 Hugging Face

Model artifacts are automatically deployed to Hugging Face Hub via GitHub Actions.