categorization power of ontologies with respect to focus classes
TRANSCRIPT
Categorization Power of Ontologies with Respect to
Focus ClassesVojtěch Svátek, Ondřej Zamazal, Miroslav Vacura
University of Economics, PragueCzech Republic
Warning: slides go beyond the paper content!
„Categorization“?
„Categorization“?
• Not meant as DL TBox classification…
C ≡ CExp 1
D ≡ CExp 2
C ⊑ D
„Categorization“?
• Not meant as DL TBox classification…
C ≡ CExp 1
D ≡ CExp 2
C ⊑ D
„Categorization“?
• Not meant as DL TBox classification…
• … but as assignment of individuals to concepts (ABox classification)
C ≡ CExp 1
D ≡ CExp 2
C ⊑ D
I J K
CExp 1 CExp 2 CExp 3
I ∈CExp 1
J ∈CExp 2
K ∈CExp 3
„Categorization“?
• Not meant as DL TBox classification…
• … but as assignment of individuals to concepts (ABox classification) – incl. compound CExp’s
C ≡ CExp 1
D ≡ CExp 2
C ⊑ D
I J K
CExp 1 CExp 2 CExp 3
I ∈CExp 1
J ∈CExp 2
K ∈CExp 3
Categorization power?
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification in a specific task
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification in a specific task
• … but as• Task-independent characteristic of an ontology
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification in a specific task
• … but as• Task-independent characteristic of an ontology
• Function of the number of categories expressible using this ontology
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification in a specific task
• … but as• Task-independent characteristic of an ontology
• Function of the number of categories expressible using this ontology
• Categories correspond to some kind of CExp’s over the ontology signature
Categorization power?
• Not meant as classification accuracy or similar…• Characteristic of a model / algorithm
• Measured as capability of automatically performing correct classification in a specific task
• … but as• Task-independent characteristic of an ontology
• Function of the number of categories expressible using this ontology
• Categories correspond to some kind of CExp’s over the ontology signature
Why „focused“?
Why „focused“?
• We are rarely completely ignorant about the initial class of individual/s to be (further) sub-categorized
Why „focused“?
• We are rarely completely ignorant about the initial class of individual/s to be (further) sub-categorized
• Mostly there exists a • initial reasonably specific named class – focus class
• …that is trivially assigned to the individual
• …prior to its non-trivial (focused) sub-categorization
Why „focused“?
• We are rarely completely ignorant about the initial class of individual/s to be (further) sub-categorized
• Mostly there exists a • initial reasonably specific named class – focus class
• …that is trivially assigned to the individual
• …prior to its non-trivial (focused) sub-categorization
Why „focused“?
• We are rarely completely ignorant about the initial class of individual/s to be (further) sub-categorized
• Mostly there exists a • initial reasonably specific named class – focus class
• …that is trivially assigned to the individual
• …prior to its non-trivial (focused) sub-categorization
Even if exceptions exist…
Even if exceptions exist…
What is the focus class for categorization?
Even if exceptions exist…
X ∈ Thing
Even if exceptions exist…
X ∈ Thing
Not “reasonably specific”
Even if exceptions exist…
X ∈ CelestialBody⊔ Insect⊔ Aircraft⊔ EyeDisorder⊔ BrainDisorder⊔ …
Even if exceptions exist…
X ∈ CelestialBody⊔ Insect⊔ Aircraft⊔ EyeDisorder⊔ BrainDisorder⊔ …
Compound, not named
Even if exceptions exist…
…in most cases we start withalready some „light“
Helicopter icon made by Freepik from www.flaticon.com
…in most cases we start withalready some „light“
Helicopter icon made by Freepik from www.flaticon.com
X ∈ Helicopter
…in most cases we start withalready some „light“
Helicopter icon made by Freepik from www.flaticon.com
X ∈ Helicopter
“Reasonably specific” named class
…in most cases we start withalready some „light“
Helicopter icon made by Freepik from www.flaticon.com
X ∈ Helicopter
Focus class
What CExp’s are “categories”? Do compound ones really count?
What CExp’s are “categories”? Do compound ones really count?
• Named subclasses of focus class are primordialcontributors to „focused categorization power“ ofthe ontology
What CExp’s are “categories”? Do compound ones really count?
• Named subclasses of focus class are primordialcontributors to „focused categorization power“ ofthe ontology
• Hypothesis 1: Some compound CExp’s should also count; such that a human ontologist would be willing to transform to named classes
What CExp’s are “categories”? Do compound ones really count?
• Named subclasses of focus class are primordialcontributors to „focused categorization power“ ofthe ontology
• Hypothesis 1: Some compound CExp’s should also count; such that a human ontologist would be willing to transform to named classes• proposed term: ontologistic category
What CExp’s are “categories”? Do compound ones really count?
• Named subclasses of focus class are primordialcontributors to „focused categorization power“ ofthe ontology
• Hypothesis 1: Some compound CExp’s should also count; such that a human ontologist would be willing to transform to named classes• proposed term: ontologistic category
• Hypothesis 2: The likelihood of a CExp to be considered as ontologistic category is correlated with its internal structure
What CExp’s are “categories”? Do compound ones really count?
What CExp’s are “categories”? Do compound ones really count?
Celestial Emporium of Benevolent Knowledgeby J. L. Borges in essay "The Analytical Language of John Wilkins (1942)
ako sub-categorization of focus class Animal:• Those that belong to the emperor• Embalmed ones• Those that are trained• Sucking pigs• Mermaids (or Sirens)• Fabulous ones• Stray dogs• Those that are included in this classification• Those that tremble as if they were mad• Innumerable ones• Those drawn with a very fine camel hair brush• Et cetera• Those that have just broken the flower vase• Those that, at a distance, resemble flies
What CExp’s are “categories”? Do compound ones really count?
Celestial Emporium of Benevolent Knowledgeby J. L. Borges in essay "The Analytical Language of John Wilkins (1942)
ako sub-categorization of focus class Animal:• Those that belong to the emperor• Embalmed ones• Those that are trained• Sucking pigs• Mermaids (or Sirens)• Fabulous ones• Stray dogs• Those that are included in this classification• Those that tremble as if they were mad• Innumerable ones• Those drawn with a very fine camel hair brush• Et cetera• Those that have just broken the flower vase• Those that, at a distance, resemble flies
Not a valid systematic categorization by any means…
What CExp’s are “categories”? Do compound ones really count?
Celestial Emporium of Benevolent Knowledgeby J. L. Borges in essay "The Analytical Language of John Wilkins (1942)
ako sub-categorization of focus class Animal:• Those that belong to the emperor• Embalmed ones• Those that are trained• Sucking pigs• Mermaids (or Sirens)• Fabulous ones• Stray dogs• Those that are included in this classification• Those that tremble as if they were mad• Innumerable ones• Those drawn with a very fine camel hair brush• Et cetera• Those that have just broken the flower vase• Those that, at a distance, resemble flies
Not a valid systematic categorization by any means… but can individualcategories be viewed as ontologistic?
What CExp’s are “categories”? Do compound ones really count?
Celestial Emporium of Benevolent Knowledgeby J. L. Borges in essay "The Analytical Language of John Wilkins (1942)
ako sub-categorization of focus class Animal:• Those that belong to the emperor• Embalmed ones• Those that are trained• Sucking pigs• Mermaids (or Sirens)• Fabulous ones• Stray dogs• Those that are included in this classification• Those that tremble as if they were mad• Innumerable ones• Those drawn with a very fine camel hair brush• Et cetera• Those that have just broken the flower vase• Those that, at a distance, resemble flies
Intuitive rewriting to DL CExp’s
• “Those that are trained”∃ trainedBy . ⊤
• “Those that have just broken the flower vase”∃ broke . Vase
• “Those that belong to the emperor”∃ belongsTo . { Emperor }
Intuitive rewriting to DL CExp’s
• “Those that are trained”∃ trainedBy . ⊤
• “Those that have just broken the flower vase”∃ broke . Vase
• “Those that belong to the emperor”∃ belongsTo . { Emperor }
• This is exactly the 3 compound CExp typesconsidered in the paper, beside C (named subclass): ∃ P.⊤ ∃ P.C ∃ P.{i}
Now peeping into the paper…
Now peeping into the paper…
• Descriptive formalism considering a language Lrestricting the form of CExp’s
Now peeping into the paper…
• Descriptive formalism considering a language Lrestricting the form of CExp’s
• Axiom patterns with mapping to CExp’s• E.g.,
P rdfs:domain FC P rdfs:range D C rdfs:subClassOf D (P,C) prun3(FC) yielding∃ P.Cas category (CExp)
Now peeping into the paper…
• Descriptive formalism considering a language Lrestricting the form of CExp’s
• Axiom patterns with mapping to CExp’s• E.g.,
P rdfs:domain FC P rdfs:range D C rdfs:subClassOf D (P,C) prun3(FC) yielding∃ P.Cas category (CExp)
• Weighted sum of axiom pattern occurrence yields focused ontologistic categorization power (FOCP)
Now peeping into the paper…
• Survey on axiom pattern occurrence in two ontology collections (LOV and OntoFarm)• Pattern for ∃ P.⊤ most widespread, followed by C and
then ∃ P.C, while ∃ P.{i} is the rarest
Now peeping into the paper…
• Survey on axiom pattern occurrence in two ontology collections (LOV and OntoFarm)• Pattern for ∃ P.⊤ most widespread, followed by C and
then ∃ P.C, while ∃ P.{i} is the rarest
• Assessment of a sample of 59 compound CExp’s by human ontologists (3 experts and 27 students)• ∃ P.{i} accepted as “ontologistic” most often (appr. 70%),
followed by ∃ P.C (appr. 50%); ∃ P.⊤ was least successful (appr. 30%)
Now peeping into the paper…
• Survey on axiom pattern occurrence in two ontology collections (LOV and OntoFarm)• Pattern for ∃ P.⊤ most widespread, followed by C and
then ∃ P.C, while ∃ P.{i} is the rarest
• Assessment of a sample of 59 compound CExp’s by human ontologists (3 experts and 27 students)• ∃ P.{i} accepted as “ontologistic” most often (appr. 70%),
followed by ∃ P.C (appr. 50%); ∃ P.⊤ was least successful (appr. 30%)
• Interesting insights drawn from disagreement cases(cf. Chris’ keynote)
Hypotheses revisited
• Hypothesis 1: Some compound CExp’s should also count; such that a human ontologist would be willing to transform to named classes
• Hypothesis 2: The likelihood of a CExp to be considered as ontologistic category is correlated with its internal structure
Hypotheses revisited
• Hypothesis 1: Some compound CExp’s should also count; such that a human ontologist would be willing to transform to named classes
• Hypothesis 2: The likelihood of a CExp to be considered as ontologistic category is correlated with its internal structure
• Both supported by the human-centric experiment (even if the sample is small)
Examples of CExp’s
• Best rated• Focus class ofrd:FridgeFreezer
category ofrd:styleOfUnit value ofrd:SingleDoori.e. type ∃ P.{i}
Examples of CExp’s
• Best rated• Focus class ofrd:FridgeFreezer
category ofrd:styleOfUnit value ofrd:SingleDoori.e. type ∃ P.{i}
• Worst rated• Focus class gr:DayOfWeek
category gr:hasNext value gr:Fridayi.e. also type ∃ P.{i}
• Focus class sigkdd:Conferencecategory sigkdd:City of conference some Thingi.e. type ∃ P.⊤
Any practical use foreseen?
Any practical use foreseen?
• Entity reuse in ontologies/vocabularies and linkeddataset schemas• Reuse strategies are now a widely researched topic in LD
• Estimated FOCP can serve as additional criterion to those already described in literature
Any practical use foreseen?
• Entity reuse in ontologies/vocabularies and linkeddataset schemas• Reuse strategies are now a widely researched topic in LD
• Estimated FOCP can serve as additional criterion to those already described in literature
• Entity transformation• Some applications may only allow to pick named classes
in an entity categorization task
• Unnamed ontologistic categories can be transformed to named ones so as to become considered by these applications
Future directions
Future directions
• Test further CExp’s and corresponding patterns, e.g. possible addition of owl:inverseOf
Future directions
• Test further CExp’s and corresponding patterns, e.g. possible addition of owl:inverseOf
• Use of linguistic cues aside the logical patterns
Future directions
• Test further CExp’s and corresponding patterns, e.g. possible addition of owl:inverseOf
• Use of linguistic cues aside the logical patterns
• Combine the expertise-based approach to “ontologistic status” prediction for CExp’s with an empirical approach, based on their population with instances
Future directions
• Test further CExp’s and corresponding patterns, e.g. possible addition of owl:inverseOf
• Use of linguistic cues aside the logical patterns
• Combine the expertise-based approach to “ontologistic status” prediction for CExp’s with an empirical approach, based on their population with instances
• Implementation of the approach in tools
Thanks! Questions?