A Practical Introduction to Ontologies & OWL Session 1: Primitive Classes in OWL Nick Drummond & Matthew Horridge

Overview Pizzas – Card Sorting Protégé Introduction Subsumption Creating a Class Hierarchy Consistency Disjointness Relationships & Properties Restrictions Polyhierarchies - Issues

Our Domain Pizzas have been used in Manchester tutorials for years. Pizzas were selected as a domain for several reasons: They are fun They are internationally known They are highly compositional They have a natural limit to their scope They are fairly neutral Although arguments still break out over representation Even pizzas can do this - its an inevitable part of knowledge modelling ARGUING IS NOT BAD!

You are the Expert Most often it is not the domain expert that formalises their knowledge – because of the complexity of the modelling it is normally a specialist “knowledge engineer” Hopefully, as tools get easier to use, this will change Having access to experts is critical for most domains Luckily, we are all experts in Pizzas, so we just need some material to verify our knowledge…

Reference Materials Having references to validate decisions, and act as provenance can be useful for maintaining an ontology Mistakes, omissions and intentions can be more easily traced if a reference can be made When building, we highly recommend documenting your model as you go – keeping provenance information is a good way of doing this We have provided you with a pizza menu and several cards with ingredients on

Our Ontology When building an ontology we need an application in mind – ontologies should not be built for the sake of it Keep the application in mind when creating concepts – this should help you scope the project The PizzaFinder application has been developed so that you can plug your ontology in at the end of the day and see it in action Let us know your ideas for extending the application

Our Application www.co-ode.org/downloads/pizzafinder/

Exercise 1: Card Sorting

Card Sorting - Issues different viewpoints Ambiguity Missing Knowledge Tomato – Vegetable or Fruit? culinary vs biological Ambiguity words not concepts Missing Knowledge What is peperonata? multiple classifications (2+ parents) lots of missing categories (superclasses?) competency questions What are we likely to want to “ask” our ontology? bear the application in mind

OWL Constructs Person Country Class (concept) Individual (instance) lives_in Person Country Elvis Hai Holger Kylie S.Claus Belgium Paraguay China Latvia lives_in Class (concept) lives_in Individual (instance) has_pet has_pet arrow = relationship label = Property Rudolph Flipper has_pet Animal

OWL Constructs: Classes Eg Mammal, Tree, Person, Building, Fluid, Company Classes are sets of Individuals aka “Type”, “Concept”, “Category” Membership of a Class is dependent on its logical description, not its name Classes do not have to be named – they can be logical expressions – eg things that have colour Blue A Class should be described such that it is possible for it to contain Individuals (unless the intention is to represent the empty class) Classes that cannot possibly contain any Individuals are said to be inconsistent

OWL Constructs: Properties Eg hasPart, isInhabitedBy, isNextTo, occursBefore Properties are used to relate Individuals We often say that Individuals are related along a given property Relationships in OWL are binary: Subject  predicate  Object Individual a  hasProperty  Individual b nick_drummond  givesTutorial  Manchester_ProtegeOWL_tutorial_29_June_2005 N-ary relationships can be modelled with workarounds in OWL, but this design pattern will not be discussed today

OWL Constructs: Individuals Eg me, you, this tutorial, this room Individuals are the objects in the domain aka “Instance”, “Object” Individuals may be (and are likely to be) a member of multiple Classes

… Is a knowledge modelling environment Is free, open source software Is developed by Stanford Medical Informatics Has a large user community (approx 30k)

… core is based on Frames (object oriented) modelling has an open architecture that allows other modelling languages to be built on top supports development of plugins to allow backend / interface extensions now supports OWL through the Protégé-OWL plugin So let’s have a look…

Protégé-OWL

Class Hierarchy Subsumption hierarchy Structure as asserted by the ontology engineer owl:Thing is the root class

Subsumption Superclass/subclass relationship, “isa” All members of a subclass can be inferred to be members of its superclasses owl:Thing: superclass of all OWL Classes A A subsumes B A is a superclass of B B is a subclass of A All members of B are also members of A B

Class Editor Class annotations (for class metadata) Class name and documentation Properties “available” to Class Disjoints widget Conditions Widget Class-specific tools (find usage etc)

Exercise 2: Create Class Hierarchy Before you start: If you are borrowing a laptop, please take a note of the number on the top cover Please save all files to: My Documents\ontologies Because you will be needing your work again tomorrow

Exercise 2: Create Class Hierarchy You will notice that we use naming conventions for our ontology entities Typically, we use CamelNotation with a starting capital for Classes Use whatever conventions you like It is helpful to be consistent – especially when trying to find things in your ontology

Saving OWL Files OWL = easy to make mistakes = save regularly Select File  Save Project As A dialog (as shown) will pop up Select a file directly by clicking the button on the top right You will notice that 2 files are created .pprj – the project file this just stores information about the GUI and the workspace .owl – the OWL file this is where your ontology is stored in RDF/OWL format Select OK

Loading OWL files If you only have an OWL file: - File New Project - Select OWL Files as the type - Tick Create from existing sources - Next to select the .owl file If you’ve got a valid project file*: - File  Open Project - select the .pprj file * ie one created on this version of Protégé - the s/w gets updated once every few days, so don’t count on it unless you’ve created it recently– safest to build from the .owl file if in doubt

Consistency Checking We’ve just created a class that doesn’t really make sense What is a MeatyVegetableTopping? We’d like to be able to check the logical consistency of our model This is one of the tasks that can be done automatically by software known as a Reasoner Being able to use a reasoner is one of the main advantages of using a logic-based formalism such as OWL (and why we are using OWL-DL)

Reasoners Reasoners are used to infer information that is not explicitly contained within the ontology You may also hear them being referred to as Classifiers Standard reasoner services are: Consistency Checking Subsumption Checking Equivalence Checking Instantiation Checking Reasoners can be used at runtime in applications as a querying mechanism (esp useful for smaller ontologies) We will use one during development as an ontology “compiler”. A well designed ontology can be compiled to check its meaning is that intended

Reasoners and Protégé Protégé-OWL supports the use of reasoners implementing the DIG interface This means that the reasoner you choose is independent of the ontology editor, so you can choose the implementation you want depending on your needs (eg some may be more optimised for speed/memory, others may have more features) These reasoners typically set up a service running locally or on a remote server – Protégé-OWL can only connect to reasoners over an http:// connection

Accessing the Reasoner Classify taxonomy (and check consistency) Compute inferred types (for individuals) Just check consistency (for efficiency)

Reasoning about our Pizzas When we classify an ontology we could just use the “Check Consistency” button but we’ll get into the habit of doing a full classification as we’ll be doing this later The reasoner dialog will pop up while the reasoner works When the reasoner has finished, you will see an inferred hierarchy appear, which will show any movement of classes in the hierarchy If the reasoner has inferred anything about our model, this is reported in the reasoner dialog and in a separate results window inconsistent classes turn red moved classes turn blue

Disjointness OWL assumes that classes overlap MeatTopping VegetableTopping = individual This means an individual could be both a MeatTopping and a VegetableTopping at the same time We want to state this is not the case

Disjointness If we state that classes are disjoint MeatTopping VegetableTopping = individual This means an individual cannot be both a MeatTopping and a VegetableTopping at the same time We must do this explicitly in the interface

ClassesTab: Disjoints Widget Add siblings as disjoint Add new disjoint Remove disjoint siblings List of disjoint classes

Exercise 3: Add Disjoints Run a reasoner locally on your machine Your reasoner may be found at…

Why is MeatyVegetableTopping Inconsistent? We have asserted that a MeatyVegetableTopping is a subclass of two classes we have stated are disjoint The disjoint means nothing can be a MeatTopping and a VegetableTopping at the same time This means that MeatyVegetableTopping can never contain any individuals The class is therefore inconsistent This is what we expect! It can be useful to create classes we expect to be inconsistent to “test” your model – often we refer to these classes as “probes” – generally it is a good idea to document them as such to avoid later confusion

Other Inconsistencies? Your ontology is likely to have several classes with multiple parents We call this a tangle As we have seen, a class cannot have 2 disjoint parents – it will be inconsistent To remove other inconsistencies you will have to be careful about where your disjoints are – remove disjoints between multiple parents by hand This is obviously an awkward thing to manage – we will later show you how to manage your tangle to simplify these issues

What have we got? We’ve created a tangled graph (not a tree – multiple parents) of mostly disjoint classes Disjoints are inherited down the subsumption hierarchy eg something that is a TomatoTopping cannot be a Pizza because its superclass, PizzaTopping, is disjoint from Pizza You should now be able to select every class and see its siblings in the disjoints widget (if it has any)

What are we missing? This is not a semantically rich model Apart from “is kind of” (subsumption) and “is not kind of” (disjoint), we currently don’t have any other information of interest We want to say more about Pizza Individuals, such as their relationship with other Individuals Pizza PizzaTopping = individual

Relationships in OWL In OWL-DL, relationships can only be formed between Individuals or between an Individual and a data value. (In OWL-Full, Classes can be related, but this cannot be reasoned with) Relationships are formed along Properties We can restrict how these Properties are used: Globally – by stating things about the Property itself Or locally – by restricting their use for a given Class

OWL Properties Object Property – relates Individuals Datatype Property – relates Individuals to data (int, string, float etc) Annotation Property – for attaching metadata to classes, individuals or properties

Properties Tab: Property Browser Note that Properties can be in a hierarchy, although we are not going to be using this feature today

Creating Properties Delete Property New Object Property: Associates an individual to another individual not used today: - New Datatype Property (String, int etc) - New Annotation Properties for metadata - New SubProperty – ie create “under” the current selection

Creating Properties We tend to name properties using camelNotation with a lowercase letter to begin We often create properties using 2 standard naming patterns: has… (eg hasColour) is…Of (eg isTeacherOf) or other suffixes (eg …In …To) This has several advantages: It is easier to find properties It is easier for tools to generate a more readable form (see tooltips on the classes in the hierarchy later) Inverses properties typically follow this pattern eg hasPart, isPartOf Our example hasBase fits into this (we will not create the inverse in this tutorial)

Associating Properties with Classes We now have a property we want to use to describe Pizza individuals To do this, we must go back to the Pizza class and add some further information This comes in the form of Restrictions We create Restrictions using the Conditions widget Conditions can be any kind of Class – you have already added Named superclasses in the Conditions Widget. Restrictions are a type of Anonymous Class

Conditions Widget Conditions asserted by the ontology engineer Add different types of condition Definition of the class (later) Description Conditions inherited from superclasses

Logical (Anonymous) Classes Conditions Types Logical (Anonymous) Classes Create Restriction (next) Add Named Superclass Create Class Expression

Creating Restrictions Type Restricted Property Filler Expression Syntax check Expression Construct Palette

What does this mean? We have created a restriction:  hasBase PizzaBase on Class Pizza as a necessary condition hasBase Pizza PizzaBase hasBase hasBase hasBase “If an individual is a member of this class, it is necessary that it has at least one hasBase relationship with an individual from the class PizzaBase” “Every individual of the Pizza class must have at least one base from the class PizzaBase”

What does this mean? We have created a restriction:  hasBase PizzaBase on Class Pizza as a necessary condition hasBase Pizza PizzaBase hasBase “There can be no individual, that is a member of this class, that does not have at least one hasBase relationship with an individual from the class PizzaBase” hasBase hasBase

Why? We have created a restriction:  hasBase PizzaBase on Class Pizza as a necessary condition hasBase  hasBase PizzaBase hasBase hasBase PizzaBase hasBase hasBase hasBase Each Restriction or Class Expression describes the set of all individuals that satisfy the condition

Why? Necessary conditions We have created a restriction:  hasBase PizzaBase on Class Pizza as a necessary condition hasBase hasBase Pizza hasBase PizzaBase  hasBase PizzaBase hasBase hasBase hasBase Each necessary condition on a class is a superclass of that class ie The restriction  hasBase PizzaBase is a superclass of Pizza As Pizza is a subclass of the restriction, all Pizzas must satisfy the restriction that they have at least one base from PizzaBase

Exercise 4: Properties & Restrictions

Restriction Types       Existential, someValuesFrom “Some”, “At least one”  Universal, allValuesFrom “Only”  hasValue “equals x”  Cardinality “Exactly n”  Max Cardinality “At most n”  Min Cardinality “At least n”

Primitive Classes All classes in our ontology so far are Primitive We describe primitive pizzas Primitive Class = only Necessary Conditions They are marked as plain orange circles in the class hierarchy We condone building a disjoint tree of primitive classes

Polyhierarchies By the end of this tutorial we intent to create a VegetarianPizza Some of our existing Pizzas should be types of VegetarianPizza However, they could also be types of SpicyPizza or CheeseyPizza We need to be able to give them multiple parents in a principled way We could just assert multiple parents like we did with MeatyVegetableTopping (without disjoints) BUT…

Asserted Polyhierarchies We believe asserting polyhierarchies is bad We lose some encapsulation of knowledge Why is this class a subclass of that one? Difficult to maintain Adding new classes becomes difficult because all subclasses may need to be updated Extracting from a graph is harder than from a tree let the reasoner do it!

Summary You should now be able to: extract Knowledge (and act as an expert) identify components of the Protégé-OWL Interface create Primitive Classes create Properties create some basic Restrictions on a Class using Existential qualifiers