Collaborative filtering recommendation engine

Simple Collaborative recommendation engine model for product similarity estimation.

There are two endpoints:

• Item based collaborative filtering: the final model is based on item cosine distances;
• User based collaborative filtering: the final model is based on item euclidean distance.

Can be deployed using the Docker image.

Requirements

Running with Python only

Built with Python 3.7, but should work with other > Python 3.5 versions.

Install the requirements:

pip install -r requirements.txt

And run the in the /app folder:

python main.py

Running with a docker container

Requires that you have setup Docker and running.

Simple run the command to build and deploy:

docker build -t recommender-image . 
docker run -p 5000:80 --name recommender recommender-image 

Testing the application

There are two endpoints:

• /predict-tag/ to obtain the TOP 10 similar tags, given another tag;
• /predict-user/ to obtain the TOP 10 user recommended tags given a user id.

Do a GET petition to the local URL through browser with the user_id/tag_id as param:

http://localhost:5000/predict-tag/<tag_id>
http://localhost:5000/predict-user/<user_id>

Example: http://localhost:5000/predict-tag/ff0d3fb21c00bc33f71187a2beec389e9eff5332

Or using curl:

curl http://lhttp://localhost:5000/predict-tag/ff0d3fb21c00bc33f71187a2beec389e9eff5332

You should see a response:

{
    "predicted_tags":
    [
        "7ee223009403f7450993fe5d79516f1fc841e75e",
        "6b0cd6a8094daf42e766ea257a2af3571831bb32",
        "bdf147e99ee57500eb2dabcbf3cfa24e1daef357",
        "340f1eaf7ad0c07f1491338ab68cbcab30c315ec",
        "c093b1743115b3f9d368b2f7bdf54f367afccc7c",
        "61bc35a6401829bd28a8da47a2f235944ba8d2df",
        "85ef93bda0f7fb6327bd1b5ad44da26246b4360d",
        "dd3c8fd58366b577ce6b1d0f435602f11671c3dc",
        "551ec41539d9fb71200d18ec7903b1039cde594f",
        "ccc01cd0dd0becfcb86471efa1202f4a6c845545"
    ]
}