12 Node classification and link prediction with GNNs
This chapter covers
- Using graph neural networks in real-world scenarios
- Building a node classification system
- Building a link prediction system
This chapter explores how to use graph neural networks (GNNs) for node classification and link prediction. These tasks represent fundamental challenges in graph-based machine learning (ML) and are central to many real-world applications.
First we’ll discuss the application of GNNs for node classification, with a focus on anti-money laundering (AML) applications. By representing financial transactions as a graph, GNNs can be used to identify suspicious patterns, classify nodes as licit or illicit, and aid in combating financial fraud. Then we focus on link prediction in recommendation systems. We will use a GNN-based approach to predict potential user–movie interactions based on ratings. By learning embeddings for users and movies and using the links between them, we can recommend movies to users based on their preferences. Despite their different tasks and application domains, we can address both scenarios with same end-to-end framework, illustrated in figure 12.1.