This article describes how to extend the simplest formulation of Graph Neural Networks (GNNs) to encode the structure of multi-relational data, such as Knowledge Graphs (KGs).

The article includes 4 main sections:

• an introduction to the key idea of multi-relational data, which describes the peculiarity of KGs;
• a summary of the standard components included in a GNN architecture;
• a description of the simplest formulation of GNNs, known as Graph Convolutional Networks (GCNs);
• a discussion on how to extend the GCN layer in the form of a Relational Graph Convolutional Network (R-GCN) to encode multi-relational data.

#graph-neural-networks #knowledge-graph #gnns-series #hands-on-tutorials #numpy