The Data

To build out our recipe recommendation system, we used an existing dataset of food.com reviews called food-com-recipes-and-user-interactions. After cleaning the dataset and parsing it into a graph data structure, we were left with the following nodes/edges:

This data was then uploaded to TigerGraph, a graph database that allows for fast and scalable storage of graph databases. Their schema system was particularly useful for this project, as it allowed us to easily define the heterogeneous structure of the graph directly inside the database.

We then downloaded the data into a Python environment and trained a variety of graph-based models.