🎬 IMDB Sentiment Analysis (NLP)

This project builds a binary sentiment classifier (positive/negative) for IMDB movie reviews using TF-IDF and linear classifiers (Logistic Regression, Linear SVM, ComplementNB). It selects the best model via 5-fold cross-validation, saves it, and provides a desktop GUI with CustomTkinter.

⚙️ Requirements

• 🐍 Python 3.10+
• 🪟 Windows PowerShell (commands below assume Windows)

Setup (recommended: virtual environment):

cd C:\Users\Konyar\Desktop\Code\IMDB_Sentiment_Analysis_NLP
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install --upgrade pip
pip install -r requirements.txt

📦 Dataset

• 📁 Location: datasets/movie.csv
• 🧾 Expected schema: text,label
  • text: review text
  • label: 0 (negative), 1 (positive)
• 🔗 Source: Kaggle – IMDB Movie Ratings Sentiment Analysis

🧠 Train the Model

The command below trains models, selects the best via CV, evaluates on a test split, and saves the model.

python IMDB_Sentiment_Analyser.py train --csv datasets\movie.csv --model models\sentiment_pipeline.joblib

📊 What you will see

• ✅ Test Accuracy: ... → test accuracy
• 📄 Classification Report: → precision / recall / F1
• 🔢 Confusion Matrix: printed in console
• 🖼️ Confusion matrix image saved to models/confusion_matrix.png
• 💾 Model saved to models/sentiment_pipeline.joblib

🖥️ Launch the GUI

After training:

python IMDB_Sentiment_Analyser.py gui --model models\sentiment_pipeline.joblib

Enter a review and click "Predict". The GUI shows prediction (Positive/Negative) and a confidence percentage.

📝 Notes

• ⏱️ Cross-validation over ~40k rows can take a few minutes.
• 📈 Confidence comes from predict_proba if available (Logistic Regression). For margin-based models (LinearSVC), a sigmoid mapping of the decision score is shown for readability. This is not a calibrated probability but useful as a relative confidence.

🗂️ Project Structure

IMDB_Sentiment_Analysis_NLP/
  IMDB_Sentiment_Analyser.py      # Training, evaluation (with CM plot), GUI
  datasets/
    movie.csv                     # Dataset (text,label)
  models/
    sentiment_pipeline.joblib     # (Created after training)
    confusion_matrix.png          # (Created after training)
  requirements.txt
  README.md

📜 License

This project is for educational purposes. Dataset usage terms belong to their respective source.