What ontologies are

What are ontologies, what are they for, and which tools exist for handling them?

Ontologies are dynamic representations of a particular domain of interest.

To be defined as such, ontologies must have three indispensable characteristics:

• they must be explicit,
• they must be formal,
• they must be shared.

This lexical, mathematical and social ambivalence makes ontologies precious tools for very many activities of great value to a company.

From a lexical point of view, to be made explicit, ontologies must be expressed in a comprehensible language. Vocabularies, thesauri, classifications, folksonomies, and dictionaries of synonyms and antonyms all provide valid tools for making the terms of an ontology explicit (for example, “man” belongs to the species “homo sapiens”).

From a mathematical point of view, ontologies are first-order axiomatic theories that can be expressed with the axioms and theorems of description logic. In this way it is possible to write algorithms capable of processing ontologies in order to build logical deductions. This characteristic has made them an indispensable tool in the field of Artificial Intelligence.

Finally, and no less importantly, the social aspect determines the real practical usefulness of an ontology: only if it is recognised by a community — whether wide or narrow — can it have any usefulness. Ignoring the social aspect of ontologies inevitably leads to conflicts, because different communities might associate profoundly different meanings with the same word (that is, refer to different ontologies).

The last aspect concerns the dynamic nature of ontologies. Ontologies evolve continuously: some terms are added, others fall into disuse, others still change meaning — the term “IoT”, for example, did not exist a few years ago. Ontologies therefore have to be maintained over time.

What ontologies are for

One of the typical problems solved thanks to ontologies is identifying, within a set of content, the items connected to a given topic (for example, “Industry 4.0”). To do this it is necessary to identify, in the text of the content, one or more topics present in a shared vocabulary. Ontologies allow exactly that vocabulary to be defined, together with a series of relations that a program can then analyse automatically.

For example, suppose it is decided that an article is about a sensor. There is a public ontology in which the term “sensor” is defined as a component of a “device”, which is in turn a component of “IoT”, which is in turn a term that can be used in the topic “Industry 4.0”.

Figure: from a sensor to the topic “Industry 4.0” through a shared ontology.

A program capable of understanding that ontology could be able to deduce that the article about the sensor might be interesting to whoever has expressed an interest in the topic “Industry 4.0”.

Which tools exist for handling ontologies

For more than ten years the standards of the Semantic Web have defined specifications that allow ontologies to be written using a comprehensible language which is at the same time formal (that is, transformable into a theorem).

Sharing, on the other hand, is left to the web itself, which over these years has produced a multitude of ontologies in very many domains of knowledge. Some of those ontologies were pre-existing thesauri and classifications that have been formalised; very many others were created to meet the needs of individual communities or applications. For example, schema.org is an ontology shared by Google, Microsoft and Yahoo for tagging web pages according to the needs of their search engines.

The W3C has standardised various languages with which to describe ontologies — more or less expressive, more or less easy to compute in different contexts. It is precisely this standardisation that has produced a rich series of tools for editing, visualising and sharing ontologies; and in turn, the availability of tools has produced a large quantity of modular, reusable ontologies.

Why ontologies have value for companies

Every company has its own language, and therefore needs its own ontologies. But no company is an island, and it is therefore essential that, as far as possible, its ontologies are shared — or at least have a clear relationship with the ontologies of its stakeholders.

When describing ontologies formally, it is therefore good practice to refer as much as possible to standard ontologies. This operation has three values:

• it increases awareness of the specific features of the company;
• it grows the company’s culture on a specific domain by incorporating what others have already done;
• it enables the use of Artificial Intelligence algorithms.

For example, if one wants to tag one’s content, the activities will consist of:

• interviews to determine the “topics”,
• analysis of the existing ontologies,
• formalisation of the content ontologies,
• set-up of the tools for editing and visualising the ontologies,
• tagging a set of sample content using a mix of automation and manual verification activities.

The tagged content and the ontologies produced can then be used as learning models to tag all the content through machine-learning algorithms.