What is OWL (Web Ontology Language)?

Most knowledge graphs start out storing and retrieving facts. OWL is what you add when you want the graph to reason about them as well. Once the rules of a domain are written down in OWL, a reasoner can classify records, catch contradictions and fill in implied facts, and it keeps doing so as the data changes.

Take a bank. Its ontology might define a High-Risk Client as any Client who is a politically exposed person, or who has transactions above a set threshold in a high-risk jurisdiction. The reasoner checks every client against that definition and flags new ones as they qualify. Nobody has to write or maintain a separate script for the rule, because it lives in the ontology next to the data it applies to.

OWL also lets different organisations describe data the same way. FIBO in finance, SNOMED CT in healthcare and SSN for sensor networks are all published in OWL, so a company can map its internal data onto them instead of inventing its own terms.

OWL sits on top of RDF and RDFS. RDFS can say that one class is a subclass of another; OWL lets you define a class by its logic. You can combine classes with AND, OR and NOT, or describe a class through a restriction, for example every Employee who works for a Company located in the UK.

Properties can be given characteristics too. A property can be transitive, symmetric or functional, or declared the inverse of another. If partOf is transitive and A is part of B, which is part of C, the reasoner concludes that A is part of C.

Cardinality restrictions say how many values a property has, so an ontology can state that every Order has at least one LineItem. OWL works under an open-world assumption, though. An Order with no LineItem recorded is not an error to a reasoner, which assumes the line item exists and simply isn't known. If you want missing data flagged, that is a job for SHACL.

Reasoning over full OWL 2 DL can get slow on large datasets, so the standard also defines three lighter profiles. OWL 2 EL suits very large class hierarchies such as SNOMED CT. OWL 2 QL is designed for answering queries over data kept in relational databases. OWL 2 RL maps onto rule engines, and it is the profile many triplestores implement for inference.

Finance. FIBO, the Financial Industry Business Ontology, defines financial instruments, legal entities and regulatory concepts in OWL. Banks use it as a shared vocabulary across trading, risk and reporting systems, and a reasoner can classify an instrument from its properties rather than from a code someone assigned by hand.

Healthcare. SNOMED CT is a clinical terminology with several hundred thousand concepts, written in OWL 2 EL. Its editors run a reasoner over it as they work, to check that new concepts are consistent and land in the right place in the hierarchy.

Manufacturing. A carmaker can describe vehicle configurations and option packages in an ontology. When engineers add a new package, the reasoner works out which existing configurations it fits, a check that would otherwise take a manual engineering review.