Summary
In this chapter, you have learned about graph embeddings, first learning what embedding is and that nodes, edges, and whole graphs can be vectorized independently. Focusing on node embeddings, you have then learned the principles of two algorithms included in the GDS library. Node2Vec, inspired by the world of texts, is a transductive algorithm, while GraphSAGE, a message-passing algorithm, is inductive and is able to predict the embedding of unseen nodes.
You have been able to extract node embeddings for nodes stored in Neo4j using the GDS implementation of these algorithms. In addition, you have discovered the GDS model catalog and been able to train a GraphSAGE model, store it into the GDS in-memory model catalog, and reuse it to predict new node representations.
In the coming chapters, we will use these embedding models and discover a new feature of GDS: pipelines. We will use these pipelines to train a node classification model. In the following chapter, we will...
