Customer Intent Classification in Banking Sector

This project aims to develop a customer intent classification model for the banking sector using PyTorch, BentoML, Docker, and Streamlit. The model predicts the intent behind customer service inquiries, with a total of 77 classes.

Table of Contents

• Overview
• Tech Stack
• Prerequisites
• Installation
• API Endpoints
• Usage
• Dataset
• Deployment
• Contributing
• License
• Contact

Overview

Customer service plays a crucial role in the banking industry, and efficiently categorizing customer intents can significantly enhance service quality and response times. This project leverages machine learning techniques to automatically classify customer inquiries into predefined categories, facilitating streamlined communication and support processes.

Tech Stack

• Model: PyTorch is used for developing the machine learning model to classify customer intents.
• Backend: BentoML is employed to create a RESTful API endpoint for deploying the model.
• Containerization: Docker is utilized to containerize the application for seamless deployment and scalability.
• Cloud Platform: Google Cloud Platform (GCP) serves as the cloud infrastructure for hosting and managing the deployed application.
• Front-end: Python Streamlit is employed to build an intuitive web-based user interface for interacting with the model.

Prerequisites

• Python 3.8 or higher
• Docker
• Google Cloud SDK (if deploying to GCP)

Installation

1. Clone the repository:

   git clone https://github.com/devbravo/Banking-Intent-Classifier-BE.git

2. Navigate to the project directory:

   cd Banking Intent Classifier - BE

3. Install the required dependencies:

   pip install -r requirements.txt

Usage

1. Start project

   bentoml serve src.api.service:svc

2. Build Bento

    bentoml build

3 Build the Docker container:

```bash
bentoml containerize intent-classification:latest 
```

3. Run the Docker container:

   docker run --rm -p 3000:3000 intent-classification:xqciyedlbkchfytg

4. Authenticate with Google Cloud

gcloud auth configure-docker

5. **Push image to Google Cloud

docker push gcr.io/[GC-PROJECT-ID]]/[IMAGE-NAME]:[TAG]

6. Access the Streamlit frontend by navigating to http://localhost:8080 in your web browser.

API Endpoints

The project provides two main API endpoints, which are built using BentoML and exposed for inference and feedback functionality.

1. /inference

   • Description: This endpoint performs text inference, predicting the intent behind a customer’s query. • Method: POST • Endpoint URL: /inference • Request Body:

{
"text": "Will my card still arrive this week?"
}

• Response: • On success (200 OK):

{
"predicted_intent": "card_delivery_status",
"confidence_score": 0.95,
"query_id": 123
}

• On error (400 Bad Request or 500 Internal Server Error):

{
"error": "Invalid input format"  // Example error message
}

2. /submit_feedback

- Description: This endpoint allows users to submit feedback on the predicted intent. If the intent was incorrect, the user can submit the correct intent.
- Method: POST
- Endpoint URL: /submit_feedback
- Request Body:

{
"query_id": 123,
"is_correct": false,
"corrected_intent": "activate_card"
}

• query_id: The ID of the query (provided by the /inference response).

  • is_correct: A boolean value indicating whether the predicted intent was correct or not.
  • corrected_intent: The corrected intent, required only when is_correct is false.

• Response:

  • On success (200 OK):

{
"message": "Feedback submitted successfully"
}

• On error (400 Bad Request or 500 Internal Server Error):

{
"error": "Failed to submit feedback"
}

Dataset

The dataset used for training the model should be placed in the data/ directory. You can download the dataset from [link to dataset source]. Ensure that the dataset is in the correct format as expected by the training script.

Deployment

To deploy the application on Google Cloud Platform:

1. Build the Docker container:

   docker build -t gcr.io/your-project-id/customer-intent-classification .

2. Push the container to Google Container Registry:

   docker push gcr.io/your-project-id/customer-intent-classification

3. Deploy the container to Google Cloud Run:

   gcloud run deploy --image gcr.io/your-project-id/customer-intent-classification --platform managed

4. Access the deployed application via the URL provided by Google Cloud Run.

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests to contribute to the project.

License

This project is licensed under the MIT License.

Contact

If you have any questions, feel free to reach out via [email protected].