Ontology

• An ontology language is a
  • formal language
  • with a clear meaning/semantics
• Examples: OWL, RDFS, Rules, ER, UML
• An ontology is a collection of statements in an ontology language
  • typically describing the classes/concepts/entities
  • and their attributes/relationships
• Most ontology languages have an implementation that may do one or more of:
  • verify correctness of statements (syntactically/semantically)
  • infer new statements
  • create a structure/schema for data

Overview of ontology/conceptual schema use

• Communicate knowledge in a precise way
• Describe structure/meaning of data
• Make knowledge independent/separate from code
• Data integration
• Saturation (inferring implicit knowledge)
• Abstraction
• Checking correctness (inconsitencies)

Where do Ontologies Come From?

• Similar to data:
  • Made by people
  • Made by software
  • Made from contents of documents
• However, ontologies mostly made by people (directly)
• More similar to software than data

What is Ontology Engineering?

• Set of tasks aimed at building and maintaining ontologies
• Main point is often to:
  • make explicit knowledge contained in software, humans or organizations
  • make meta-data for semi-structured data
• Generally complex task, similar in many respects to software engineering

Ontology Engineering Process

• Different types of people involved
  • Ontology expert: Knows logic and ontologies
  • Domain expert: Knows the domain
  • Data manager: Knows the organization’s data
• Need to communicate and agree

Complexities in Ontology Engineering

1. What to model?
2. How to model?
3. How to maintain the model?

Complexity 1: What to model?

• Impossible to model everything within a domain
• Impossible to capture a (part of) a domain in all its details
• Sometimes impossible for model to be completely correct wrt. a domain
• Get people to understand one another
• Understand the domain
• Understand the data
• Understand how the ontology will be used
• Which foundational assumptions should one make?
  • Are you modelling the “Real World”?
  • Or are you modelling people’s conceptualization

Scoping

• Important to clearly scope what should be modelled
• Often wise to start with something restricted, and then extend
  • Industrial Ontology Foundry (IOF) example: 20 most central terms
• Still need to have an overview of what needs to go into ontology
• Different things to model: Physical things, functional locations, social/mental constructions, etc.
• Different aspects to model for physical things:
  • Physical properties (e.g. color, weight, shape)
  • Functional properties (e.g. what can it do, what does it do now)
  • Other properties (e.g. location, types, change over time)

Competency Questions

• A competency question is a question that the ontology should be able to answer
  • Which species hunt endangered species?
  • Which alien species live in marine habitats?
• Shows the vocabulary the ontology needs to introduce
• Also shows the relationships between the terms
• A very useful tool for scoping
• Also useful tool for communication between roles

Reuse

• There are lots of existing ontologies
  • See e.g. OBO Foundry and BioPortal
• Possible to reuse all or parts of existing ontologies
• Some languages, like OWL, have special support for reuse
  • import-statements
• Artsdatabanken example (list of relevant ontologies)
• Also exists libraries of modelling patterns
  • See e.g. Ontology Design Patterns

Standards

• If no ontology to reuse exists, maybe there however exists:
  • Standardized vocabularies
  • Engineering standards
  • Taxonomies
  • Scientific classifications
  • Etc.
• This will give you a skeleton to work with
• Ensures correct foundation
• Artsdatabanken example: Taxonomies
• Aibel example: ASME

Foundational Assumptions (Ontological Commitments)

• Need to decide what things exists
• What are the topmost classes, i.e. how is reality categorized
• What determines identity
  • E.g. can objects change, or do they become new objects
  • Ship of Theseus
• Can quickly end up in trouble if different foundational assumptions are made in same ontology
• Difficult to combine ontologies with different foundational assumptions

Upper Ontologies and Methodologies for Modelling

• Easier to combine (sub)ontologies with same foundational assumptions
• Consequentially, more easy to integrate ontologies with same foundations
• Upper ontologies defines the most foundational layer in an ontology
• Most upper ontologies comes with a methodology
• Different upper ontologies exists:
  • BFO: Realist, commonly used in biomedicine
  • Dolce: Conceptualist, commonly used
  • UFO: Provides a rich methodology (OntoClean)
  • Many others…

Complexity 2: How to model?

• Decide on the best way to model the domain
• Decide on best language for that modelling
• Decide how things should be named
• Handling size and complexity of ontology

Classes vs. instances vs. properties

• Can model the same thing in many ways
• Lets find a way to model sentences of the form:
  • “A wolf is a mammal living in the forest.”
  • “A cod is a fish living in the sea.”

# Ontology
Class: Mammal
    subClassOf: Animal
Class: Fish
    subClassOf: Animal
    
Class: Wolf
    subClassOf: Mammal and habitat some Forest
    subClassOf: Mammal and habitat only Forest
    
Class: Cod
    subClassOf: Fish and habitat some Sea 
    subClassOf: Fish and habitat only Sea 

# Data
Wolf(ulv123)
habitat(ulv123, loc98)
Cod(torsk456)
habitat(torsk456, loc39)

# Use:
# Inference: ulf123 a Mammal, loc98 a Forest, etc.
# Query: Query for all Mammals, Animals, etc. directly
# Cannot query for animals of same species (every animal gets type Animal
#   SELECT * WHERE { ?x a ?Species . ?y a ?Species . }

# Ontology
:wolf :inKingdom :mammal ;
    :hasHabitat :forest .

:cod :inKingdom :fish ;
    :hasHabitat :sea .

# Data
:ulv123 :hasSpecies :wolf ;
    :habitat :loc98 .
:torsk456 :hasSpecies :cod ;
    :habitat :loc39 .

# Use:
# Inference: None (but could add e.g. hasSpecies o hasHabitat subPropertyOf: hasHabitat
#    -> :ulv123 :hasHabitat :Forest
# Query: Cannot query for all Mammals, Animals, etc. directly (but indirectly with more complex wuery)
# Can query for animals of same species
#   SELECT * WHERE { ?x :hasSpecies ?species . ?y :hasSpecies ?species . }

Classes vs. individuals (cont.)

• What does the sentence really mean: “A wolf is living in the forest.”
• E.g. if we know that :ulv123 is a wolf and has habitats :loc1, :loc2 and :loc3:
  • Do wolves only live in forests, i.e. all of the habitats are forests
  • Do wolves have at least one habitat that is forest, i.e. one of the habitats is a forest
  • Do wolves most commonly/probably live in forests (cannot be modelled directly in OWL)
• Sometimes impossible to model the domain correcly
• Might still be possible to make correct model for a concrete usecase

Decide how to model things “impossible” to model

• Example in requirements: Every artefact with weight above 1,000kg must have a weight certificate.
• Example in biology: Birds can usually fly.

Choosing Ontology Language/Fragment

• Need to decide on a language, e.g.:
  • RDF, RDFS
  • OWL 2 DL, EL, QL or RL
  • SWRL, Lore, Datalog, Prolog
  • ER, UML, ORM
  • Many others (see e.g. this list)
• Should be based on need for expressiveness, computational efficiency, surrounding frameworks, etc.
• Note that greater expressiveness also often leads to more complicated ontologies
  • Restrictions often forces discipline!

Naming

• Need to decide on naming of classes, properties, individuals
• E.g. should names be meaningfull
  • <http://artsdatabanken.no/arter/Animalia/Arthropoda/Insecta/Coleoptera/Polyphaga/Cucujiformia/Cleroidea/Dasytidae/Dasytinae/Dasytini/Dasytes/plumbeus>
• Or meaningless and generated:
  • <http://artsdatabanken.no/s6854432187631>
• DBPedia vs WikiData
• Meaningful IRIs bind meaning in the name
  • Change IRI if knowledge changes (i.e. not constant)
  • E.g. change above IRI if species is moved in the taxonomy
• One reccomendation is to use meaningless IRIs, but use labels as human-readable tags
  • Easy to translate labels to different languages
  • Never need to change IRI of class/property/individual
  • However, can be difficult to read queries (compare this to this)
• Same name often used in completely different ways accross or within organizations

Modularization

• When ontologies become complex, need to break them up into smaller pieces
• This is known as modularization
• Many ways to do this:
  • Based on domain
  • Based on layers of abstraction
  • Based logical entailment
  • Based on use
• Example: Aibel ontology (100,000+ classes, 100+ standards)

Patterns and Templates

• Already covered earlier in the lectures (and in IN3060/IN4060)
• However, an important point about OTTR templates:
  • Provides a natural form of modularization
  • Different roles are responsible for different things
  • Ontology expert: Makes foundational templates for logic, upper ontologies, etc.
  • Data expert: Makes templates matching organization’s data formats and needs
  • Domain expert: Makes instances to populate ontology

Complexity 3: How to Maintain Model?

• Changes to real world
  • New or evolving standards
  • New knowlegde
  • New objects
• Changes to use
  • New use-cases
  • New types of data
  • New rules/regulations
• Backwards compatibility
• Artsdatabanken example: Two species are the same, one species split into multiple, etc.
• Ontology debugging

Design of relational schemas

• Underlying any relational schema is a conceptual model!
• Sometimes model is made explicit (ER, UML)
• Many of the same problems apply if done via ER/UML-modelling
• Need to decide on:
  • Foundational assumptions
  • Identifiers → Keys
  • Modularization → Schemas, tables
  • Instances vs. properties → Columns vs. values

                            Country
                            
| country        | region   | 2023       | 2022       | 2021       |
|----------------|----------|------------|------------|------------|
| United States  | Americas | 26,854,599 | 22,996,100 | 23,315,081 |
| China          | Asia     | 19,373,586 | 17,734,063 | 17,734,131 |
| Japan          | Asia     | 4,409,738  | 4,937,422  | 4,940,878  |
| Germany        | Europe   | 4,308,854  | 4,223,116  | 4,259,935  |
| India          | Asia     | 3,736,882  | 3,173,398  | 3,201,471  |
| United Kingdom | Europe   | 3,158,938  | 3,186,860  | 3,131,378  |
| France         | Europe   | 2,923,489  | 2,937,473  | 2,957,880  |
| Italy          | Europe   | 2,169,745  | 2,107,703  | 2,107,703  |
| Canada         | Americas | 2,089,672  | 1,988,336  | 1,988,336  |
| Brazil         | Americas | 2,081,235  | 1,608,981  | 1,608,981  |


--

            Country

| country        | region   |
|----------------|----------|
| United States  | Americas |
| China          | Asia     |
| Japan          | Asia     |
| Germany        | Europe   |
| India          | Asia     |
| United Kingdom | Europe   |
| France         | Europe   |
| Italy          | Europe   |
| Canada         | Americas |
| Brazil         | Americas |

            GDP
            
| country        | year | GPD        |
|----------------|------|------------|
| United States  | 2023 | 26,854,599 |
| China          | 2023 | 19,373,586 |
| Japan          | 2023 | 4,409,738  |
| Germany        | 2023 | 4,308,854  |
| India          | 2023 | 3,736,882  |
| United Kingdom | 2023 | 3,158,938  |
| France         | 2023 | 2,923,489  |
| Italy          | 2023 | 2,169,745  |
| Canada         | 2023 | 2,089,672  |
| Brazil         | 2023 | 2,081,235  |
| United States  | 2022 | 22,996,100 |
| China          | 2022 | 17,734,063 |
| Japan          | 2022 | 4,937,422  |
| Germany        | 2022 | 4,223,116  |
| India          | 2022 | 3,173,398  |
| United Kingdom | 2022 | 3,186,860  |
| France         | 2022 | 2,937,473  |
| Italy          | 2022 | 2,107,703  |
| Canada         | 2022 | 1,988,336  |
| Brazil         | 2022 | 1,608,981  |
| United States  | 2021 | 23,315,081 |
| China          | 2021 | 17,734,131 |
| Japan          | 2021 | 4,940,878  |
| Germany        | 2021 | 4,259,935  |
| India          | 2021 | 3,201,471  |
| United Kingdom | 2021 | 3,131,378  |
| France         | 2021 | 2,957,880  |
| Italy          | 2021 | 2,107,703  |
| Canada         | 2021 | 1,988,336  |
| Brazil         | 2021 | 1,608,981  |