This project is a complete and ready-to-use application that classifies customer reviews into predefined categories. It leverages a fine-tuned DistilBERT model for high-accuracy predictions and features a user-friendly Streamlit frontend. The entire application is designed for seamless deployment on the Hugging Face Hub.

• Data Preprocessing: Scripts to clean and prepare raw customer review data.

• Model Training: A robust training pipeline for fine-tuning a DistilBERT model for multi-class classification.

• Model Hosting: The trained model is hosted on the Hugging Face Hub, enabling direct use without local storage.

• Comprehensive Evaluation: The evaluate.py script generates a full suite of metrics, including a confusion matrix and visualizations of model performance.

• Streamlit App: A single-file, production-ready Streamlit application that handles both the frontend and backend logic.

• Professional UI: The app includes a responsive design, clear metric explanations, and professional-looking charts.

1. Data Preparation: The prepare_data.py script cleans and preprocesses the raw data from data/sample_data.csv.

2. Model Training: The train.py script fine-tunes the DistilBERT model. The output model and tokenizer are saved locally.

3. Model Upload: The upload_to_hf.py script pushes the trained model and tokenizer to the Hugging Face Hub, making it accessible to the app.

4. Model Evaluation: The evaluate.py script evaluates the model's performance on a test set and saves the results (summary_report.json, confusion_matrix.png, etc.) to output/results/.

5. App Deployment: The app.py script serves as the main application. It loads the model directly from the Hugging Face Hub and provides a user interface for classification and metric visualization.

The application uses a simplified architecture where the Streamlit frontend directly loads and runs the model. This eliminates the need for a separate FastAPI backend and is the standard practice for deploying a single-model app on Hugging Face Spaces.

Customer-Review-Classifier/
├── data/
│   ├── sample\_data.csv           \# Raw customer review data
│   └── sample\_data\_cleaned.csv   \# Cleaned and encoded data
├── output/
│   ├── distilbert\_model/         \# Local copy of the trained model (optional)
│   ├── results/
│   │   ├── summary\_report.json   \# JSON report of all evaluation metrics
│   │   ├── f1\_scores.png         \# F1 scores visualization
│   │   ├── confidence\_distribution.png \# Confidence distribution visualization
│   │   └── confusion\_matrix.png  \# Visual representation of model confusion
├── src/
│   ├── prepare\_data.py           \# Script for data cleaning and encoding
│   ├── train.py                  \# Script for fine-tuning the model
│   ├── evaluate.py               \# Script for model evaluation and reporting
│   ├── app.py                    \# The main Streamlit application
│   └── upload\_to\_hf.py           \# Script to push the model to the Hugging Face Hub
├── requirements.txt              \# Project dependencies
├── .gitignore                    \# Specifies files to ignore in Git (e.g., large models)
└── README.md

1. Install dependencies:

   pip install -r requirements.txt

2. Prepare data:

   python src/prepare_data.py

3. Train model:

   python src/train.py

4. Evaluate model:

   python src/evaluate.py

5. Upload model to Hugging Face Hub:

   huggingface-cli login
   python src/upload_to_hf.py

6. Start the app:

   python src/app.py

7. Access the app:

   • Open http://localhost:8501 in your browser after running the command above.

• Large model files are not tracked in Git. See .gitignore.
• For custom categories, update cat_map in app.py and api.py.
• For issues or improvements, open an issue or pull request.

Author: Sayantan Ghosh (Lazycoder03)