Scaling Up Graph Neural Networks with GraphSAGE
GraphSAGE is a GNN architecture designed to handle large graphs. In the tech industry, scalability is a key driver for growth. As a result, systems are inherently designed to accommodate millions of users. This ability requires a fundamental shift in how the GNN model works compared to GCNs and GATs. Thus, it is no surprise that GraphSAGE is the architecture of choice for tech companies such as Uber Eats and Pinterest.
In this chapter, we will learn about the two main ideas behind GraphSAGE. First, we will describe its neighbor sampling technique, which is at the core of its performance in terms of scalability. We will then explore three aggregation operators used to produce node embeddings. Besides the original approach, we will also detail the variants proposed by Uber Eats and Pinterest.
Moreover, GraphSAGE offers new possibilities in terms of training. We will implement two ways of training a GNN for two tasks – node...
