GraphSAGE: Scaling up Graph Neural Networks

What do UberEats and Pinterest have in common?

They both use GraphSAGE to power their recommender systems on a massive scale: millions and billions of nodes and edges.

• 🖼️ Pinterest developed its own version called PinSAGE to recommend the most relevant images (pins) to its users. Their graph has 18 billion connections and three billion nodes.
• 🍽️ UberEats also reported using a modified version of GraphSAGE to suggest dishes, restaurants, and cuisines. UberEats claims to support more than 600,000 restaurants and 66 million users. Meanwhile, it keeps recommending me tacos that gave me food poisoning.

In this tutorial, we’ll use a dataset with 20k nodes instead of billions because Google Colab cannot handle our ambitions.

 

 

 

To finish reading, please visit source site