ramonzaca / poc-product_recommender

POC - Recomender

Proof of concept for a recommendation system of products based on the Flipkart Products dataset.

Recommendations are made based on product name and product description, using cosine distance between sentence embeddings.

Request & Response Examples

API Resources

• GET /
• POST /recommend

GET /

Example: http://localhost:5000/

Description: Response with a welcome message.

Response body:

{
    'message': "Hello! This is the recommender system"
}

POST /recommend

Example: Create – POST http://localhost:5000/recommend

Description: Gives K recommendations similar to uniq_id.

Request body:

{
    'uniq_id': "f465dfcbf48cb1e4cde0f5cd1da50aac",
}

(200) Response body:

{
    'values': [
            *product_entry_1*,
            *product_entry_2*,
            *product_entry_3*,
    ],
    'scores': [0.92, 0.92, 0.87]
}

(404) Response body:

{
    'message': 'Unable to retrieve item *definetly_not_a_uniq_id*'
}

Description of components

• app.py : Flask app as REST API.

• README.md : This.

• requirements.txt : PIP requirements for the docker.

• Dockerfile : Description of the docker.

• recommender.py : Recommendation system code.

• Dockerfile : Description of the docker.

• config.py : Configuration file.

Usage

Starting from the repo folder.

• Building the image: docker build -t recommender:latest ..
• Run the image: docker run -p 5000:5000 recommender.

Notes:

• The system can also be start locally via python app.py, if all requirements are installed.