Part 4 Machine learning on knowledge graphs

This part of the book explores how representation learning and graph neural networks can transform the static knowledge contained in graphs into dynamic, learnable features:

Neural-network-based representations capture the complexity of graph structures and their entities.
Structured information can be effectively encoded in vector spaces.
Flexible feature representations support downstream tasks, from classification to link prediction.
Interpretation of learned embeddings allows automated knowledge extraction.

Chapter 9 introduces the fundamental concepts and motivations for applying machine learning (ML) to KGs, establishing why graph-based approaches better reflect real-world dependencies and how they can be applied to computational tasks.

Chapter 10 illustrates manual and semiautomated approaches to feature engineering in graph-based ML, demonstrating how graph metrics and structural patterns can be captured and utilized.

Chapter 11 shows how graph neural networks can automatically learn optimal representations from graph structures.

Chapter 12 demonstrates these concepts in action through two real-world implementations that showcase how graph neural networks can tackle business challenges while maintaining interpretability.