Intent Prediction Model for Novo Customer Feedback

Project Structure

• app/: Contains the Flask application code
  • __init__.py: Flask application factory
  • main.py: Main Flask blueprint and routes
  • model.py: Intent classification model class
• data/: Contains the customer feedback data
  • feedback.xlsx: Excel file with customer feedback
• models/: Contains the trained model artifacts
  • tfidf_vectorizer.pkl: Trained TF-IDF vectorizer
  • intent_classifier.pkl: Trained intent classifier
  • label_binarizer.pkl: Label binarizer for intents
• train_model.py: Script to train the intent prediction model
• run.py: Entry point to run the Flask application
• requirements.txt: List of required Python packages
• README.md: Project documentation

How to Run

1. Create a new conda environment and install dependencies:

   conda create -n novo_assignment python=3.9
   conda activate novo_assignment
   pip install -r requirements.txt

2. Train the model (if not already done):

   python train_model.py

3. Run the Flask application:

   python run.py

4. Use the /predict endpoint to predict intents for customer feedback.

Notes

• Ensure feedback.xlsx is placed in the data/ directory.
• The model files (tfidf_vectorizer.pkl, intent_classifier.pkl, label_binarizer.pkl) should be in the models/ directory.

Model Evaluation

• Evaluate the model using standard metrics like accuracy, precision, recall, and F1-score.
• Use a hold-out validation set or cross-validation for reliable performance estimates.

Teaching New Intents

1. Collect new labeled data for the new intents.
2. Retrain the model using the updated dataset.
3. Save the updated model artifacts to the models/ directory.

=======