Transcript ppt
Introduction to Ontologies
ECE457 Applied Artificial Intelligence
Spring 2008
Lecture #13
Outline
Ontology
Inheritance
Russell & Norvig, sections 10.1, 10.2, 10.6
CS 886 (Prof. DiMarco)
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 2
Knowledge Base
In logic, our KB was simply a list of facts
Works because we use simple examples
Won’t work in real life
Need to structure facts in KB
Make storing, searching for and retrieving
information from KB easier
Sort facts into categories
Define relationships between facts and/or
categories
Arrange relationships hierarchically
Ontology
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 3
Ontology
Representation of concepts and relationships
between concepts
Allows representation and handling of information
about objects represented in it
Can be general or domain-specific
Reusability vs. easy of design, analysis, implementation
Four main parts
Objects
Categories
Relations
Attributes
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 4
Objects and Categories
Objects
Real-world items
Apple A42, Bob the penguin
Categories
Abstractions, groups of objects
Apples, fruits, seeds, penguins, birds,
wings, physical objects
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 5
Objects and Categories
PhysicalObjects
Fruits
Apples
Birds
Seeds
Wings
A42
ECE457 Applied Artificial Intelligence
Penguins
Bob
R. Khoury (2008)
Page 6
Relations
Binary connections
Between two objects, two categories, or an
object and a category
Typical relations
IsA: A category is a kind of another
category
InstanceOf: An object is an instance of a
category
PartOf: A category is a part of any object
that’s an instance of another category
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 7
Relations
IsA
PhysicalObjects
IsA
Fruits
Birds
PartOf
IsA
Apples
Seeds
InstanceOf
PartOf
Wings
Penguins
InstanceOf
A42
ECE457 Applied Artificial Intelligence
IsA
Bob
R. Khoury (2008)
Page 8
Relations
Objects and categories are constant symbols
in FOL
Relations are predicates in FOL
InstanceOf(A42,Apples)
IsA(Apples,Fruits)
PartOf(Seeds,Fruits)
IsA(Fruits,PhysicalObjects)
InstanceOf(Bob,Penguins)
IsA(Penguins,Birds)
PartOf(Wings,Birds)
IsA(Birds,PhysicalObjects)
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 9
Attributes
Properties of objects and categories
Intrinsic properties
Part of the very nature of the category
Boiling point, edible, can float, …
Extrinsic properties
Specific to each object
Weight, length, age, …
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 10
Attributes
PhysicalObjects
Mass=? Age=?
IsA
IsA
Fruits
Edible=Yes
IsA
Apples
Birds
PartOf
Colour={Red,Green}
Seeds
InstanceOf
A42
Kind=McIntosh
ECE457 Applied Artificial Intelligence
PartOf
Wings
Feather=Yes
IsA
Penguins
InstanceOf
Bob
Age=2 years
R. Khoury (2008)
Page 11
Attributes
Relations are functions or predicates in
FOL
Edible(Fruits)
Feather(Birds)
Mass(PhysicalObjects,x)
Age(PhysicalObjects,x)
Colour(Apples,Red) Colour(Apples,Green)
Kind(A42,McIntosh)
Age(Bob,2)
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 12
Inheritance
Passing properties from general
categories to specialized categories or
objects
Categories/objects have to be connected
Easily gain a great deal of information
about children
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 13
Inheritance Network
Fruits are edible, Fruits
apple is a fruit,
Edible=Yes
therefore apple is
edible
Birds have
Apples
feathers, penguin
is a bird, therefore
penguin has
feathers
A42
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Birds
Feather=Yes
Penguins
Bob
Page 14
Inheritance Network
Inheritance network is sentences in FOL
x IsA(x,Fruits) Edible(x)
x InstanceOf(x,y) IsA(y,Fruits)
Edible(x)
x IsA(x,Bird) HasFeathers(x)
x InstanceOf(x,y) IsA(y,Bird)
HasFeathers(x)
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 15
Inheritance Problems
Child inherits contradicting
attributes from its parent
and grandparent
Shortest path heuristic
Penguins closer than Birds
Danger: redundant links
Inferential distance
Penguins closer than Birds
because there is a path from
Bob to Birds through Penguins
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Birds
Fly=Yes
Penguins
Fly=No
Bob
Fly=?
Page 16
Inheritance Problems
Ambiguous network
Child inherits contradicting attributes from
its parents
Inferential distance doesn’t apply!
Quaker
Pacifist=Yes
Republican
Pacifist=No
Richard Nixon
Pacifist=?
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 17
Solutions to Ambiguous Nets
Credulous approach
Sceptical approach
Randomly select one value
Assign no value
Shortest path heuristic
Assign the value resulting from the
shortest path in the network
Path length not a relevance measure
Shortcuts in network
Use of many fine-grained distinctions
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 18
Ontology Learning
One of the main challenges in ontology
research today
Often done manually
Partially-automated techniques
Still need manual checking
Start from a manually-constructed core
ontology
Work best for specialized ontologies
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 19
Automated Ontology Learning
Seed ontologies
Input texts
Natural language
processing system
Knowledge
extractor
Lexicon
Inference
rules
Databases
KB manager
Ontology
Ontology
manager
KB
Engineer
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 20
Ontology Example: WordNet
English vocabulary ontology
Handles nouns, verbs, adjectives and
adverbs independently
Some call it a lexical hierarchy
Nouns ontology biggest and most used
Nouns subdivided in 25 classes
Often used to measure the
similarity/distance between words
So successful, other languages WordNet
are being created
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 21
WordNet Relations
{organism,
living thing}
{animal, fauna}
Synonymy
Hyponymy
{bird}
{robin, redbreast}
ECE457 Applied Artificial Intelligence
Sets of synonyms (synsets)
are the basic building blocks
of WordNet
Also an Antonymy relation
“is a kind of”
Hyponym(Robin,Bird)
Hypernym(Bird,Robin)
Organizes WordNet into
lexical hierarchy
R. Khoury (2008)
Page 22
WordNet Relations
{body part }
Meronymy
{external
{feature,
body part } lineament }
{face,
human face}
{bird}
{mouth}
“is a part of”, “has a”
Meronym(beak,bird)
Holonym(bird,beak)
Intertwined with
Hyponymy
{jaw}
{beak, bill, neb, nib}
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 23
WordNet Construction
Created at Cognitive Science Laboratory,
Princeton University
Started with Brown Corpus and integrated
pre-existing thesaurus
Manually created, expanded and verified
Online effort
Uses home-made programs to help
1985: started
1993: 57,000 nouns in 48,800 synsets
1998: 80,000 nouns in 60,000 synsets
2007: 117,000 nouns in 81,000 synsets
ECE457 Applied Artificial Intelligence
R. Khoury (2008)
Page 24