Ontology and knowledge graphs simply explained.

Recently, an Australian government minister authored this tweet:

“To recap, having a whole of govt architecture allows us to build an ontology of capabilities across govt. Coupled with a more agile funding model, it will allow us to move more quickly when trying new solutions and capabilities or scaling up platforms to address emerging needs.”

This tweet was met with considerable derision at the time because readers did not understand a word of it. The tweet makes sense if the reader knows what an ontology is (providing you ignore the compulsory middle-management buzz-wordage).

So, what is an ontology?

It is best to start at the start. Sir Tim Berners-Lee invented two webs, they are:

  • The web of documents which we use every day and
  • The Semantic web, a parallel web which is populated by interconnected ontologies.

The web of documents is human-readable and the Semantic web is machine-readable.

The ontology

An ontology is a machine-readable web document that is a model or a ‘snapshot’ of some bit of human reality. The reality may be your organisation, a club or a collection of medical terms. It is how machines ‘understand’ some part of our reality.

Ontologies map our reality using the relationships between things. Things may be concrete (like people) or abstract (like feelings). Things and their relationships are identified by their commonly accepted names (semantics). On the Semantic web, millions of ontologies may be connected together by relationships, the more we have, the greater the resolution of reality that machines may understand.

Just like our reality, an ontology has rules (Axioms) embedded in it which determine what a thing is and how things are related to it.

Axioms determine the nature of a thing. Say, the axioms for a dog may be that a dog has four legs and goes “woof”.  All dogs that are attached under this archetypal dog must fulfil these axioms. That is, a dog is a dog because it has four legs and goes “woof”. This ‘inheritance’ is passed down to all dogs and a dog is not a dog unless it fulfils this axiom. Inheritance is an important part of an ontology. Also, a cat has different axioms and can never be a dog.

There are plenty of ontologies which just use inheritance for collections of things, how do humans read them? Often ontologies are made to look like taxonomies, collections or lists of things. They display a top-level thing and all the things related to them by inheritance. Bioportal is a good place to start if you want to check out this type of ontology. Bioportal contains 800+ medical ontologies which are basically searchable taxonomies of terms.

Knowledge graphs

Who puts information into an ontology? The short answer is a person who knows their reality the best, a ‘domain expert’. The problem is that few domain experts are capable of constructing an ontology. This is where knowledge graphs come in. Knowledge graphs are human-readable, easily constructed by anyone and can be turned into a machine-readable ontology.

Following is a simple knowledge graph showing the relationship between Bob and Alice. A machine or human may glean the knowledge that “Bob knows Alice” but “Alice doesn’t know Bob”. A machine may be trained to detect these relationships in an ontology as ‘a one way relationship’, maybe Bob is an admirer from afar?

bob knows alice

Knowledge graphs can be very subjective if they are constructed from one person’s point of view. They are best constructed using many people’s points of view and collated.

Artificial intelligence

Some ontologies are capable of Artificial Intelligence (AI). They use the Semantic Web Rule Language (SWRL) to infer new knowledge from existing knowledge. Knowledge inference is called ‘reasoning’. This way, on ontology may become a ‘knowledge base’ for a larger AI application. A nursing example that uses patient’s functioning scales to suggest resources in an ontology may be found on my website  => demonstrations => AI Reasoning.

So, the politician’s tweet above basically says that we can have an ontology of our capabilities and maybe identify emerging needs. Either way, a map is a good thing to have in an unexplored area.

Philip Shields RN PhD