9 Machine learning on knowledge graphs: A primer approach
This chapter covers
- Understanding machine learning on knowledge graphs
- Exploring common machine learning tasks performed on graphs
- Understanding the role of node and relationship representations
Building knowledge graphs is a crucial step in developing intelligent systems. It enables us to acquire holistic knowledge from multiple and diverse data sources, representing it in a way that supports exploration, navigation, and more advanced analytics. So far, we have seen how to query a graph and extract relevant information, how to navigate through nodes and relationship types, and even how to extract statistical information to validate the import process and evaluate the “quality” of the knowledge stored in the KG. These are all important steps in building intelligent advisory systems (IASs).
In an IAS, “advising” provides insights that users cannot extract on their own. For example, how can a researcher efficiently navigate a vast KG containing diseases, proteins, genes, and compounds to identify potential opportunities for repurposing drugs? Or, how can a clinician combine patients’ symptoms and DNA sequences with literature, clinical trials, and standard protocols to develop personalized treatment plans? There are many such scenarios. And most, if not all, require using machine learning (ML) algorithms that take the knowledge in the graphs as input.