Youtube music recommender

This is a simple youtube music/video recommender. You can checkout the deployed solution on Heroku:

https://thawing-shore-99052.herokuapp.com/

The problem is solved by:

1. Scrapping video data from Youtube pages;
2. Extracting the video information from each page;
3. Preprocess data from each video into a single dataset;
4. Manually label some of samples, active learning the rest;
5. Extract features from the dataset;
6. Train a Random Forest and a LightGBM model and ensamble them;
7. Build a simple app to serve the model through Heroku.

Running:

Clone the repository and then go to /src/data_science folder.

• Scrap data from Youtube pages using:

python search_data_collection.py

• Extract information from the pages by:

python search_data_parsing.py

• Process the video data using

python video_data_processing.py

• Manually label the data creating a new column named "y".
• Fit the final model using:

python final_model.py

Now that the model is trained, we can deploy it. Go to /src/deploy folder.

Run Locally

Start the database with:

python db_starter.py

Then run the app with:

python app.py

Make a local Docker image to run

Build the docker image:

docker build . -t deploy_ytr

Run the Docker image:

docker run -e PORT=8000 -p 80:80 deploy_ytr