This is a recommener system for cooking.nytimes.com. There is a scraper to download user's favourited recipes, which is used to seed the recommender. The main app is a Flask webserver with two endpoints /addData and /predictInterests
This project uses AWS DynamoDB for its datastore, so before you start, create a user in AWS with read/write permissions for DyanamoDB, and set your environment variables AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY appropriately.
Important: Run all commands from inside the recommender directory, not the project root.
The scraper uses selenium with Firefox, to run install Firefox (tested with version 30.0, might not work with later versions.
To run the scraper, simply
python scraper.py
This will automatically create the DynamoDB tables if they don't exist already, and write the scraped values to DynamoDB
Run the webservice with
python app.py <user_table_name> <recipes_table_name>
The app has three endpoints.
This is for adding an array of user likes, eg.
curl http://127.0.0.1:5000/addData?user_id="321"&recipe_ids=["123", "456"]
This is to update the item-similarity matrix used for making predictions:
curl http://127.0.0.1:5000/refreshPredictions
This will get the top 5 recommendations for a user based on item similarity with known users, since last running refreshPredictions
curl http://127.0.0.1:5000/predictInterests?user_id=123
Run doctests for the recommender with:
python -m doctest -v recommender.py
This will create a test DynamoDB table, upload a training sample of data (200 users restricted to 100 recipes scraped from the NYTimes website), and run recommender against test subset of these, calculating the Average Precision@5 compared to the baseline recommender which recommends the top 5 recipes.
Run the integration testing script as:
python recommender_test.py