Semantic Web Technologies

The Semantic Web:Means Many Things to Many People

What is the Semantic Web?

The Semantic Web is an evolving extension of the World Wide Webin which web content can be expressed not only in natural language,but also in a form that can be understood, interpreted and used bysoftware agents, thus permitting them to find, share and integrateinformation more easily.

It derives from W3C director Tim Berners-Lee's vision of the Web asa universal medium for data, information, and knowledge exchange.

The Story

The Story Analyzed

• Agent finds hospitals based on attributes:

– Compatible insurance plan

– Compatible location

– Available appointment times

– Available services

– Required quality rating

• Requires:

– Access to hospital data

– Access to personal data (query, calendar, location)

– Data includes all attribute values

– Compare hospitals’ and query’s attribute values (unified rep)

– Explain decisions and inferences

Challenges

• Missing data

• Extra data

• Different representations of data (vocabulary):

– Different units of measure

– Different names for same entities, attributes, values

– Same names for different entities, attributes, values

– Different granularity of entity classes

Solutions

• Missing data Infer from available data.

• Extra data Filter using relevance.

• Different representations of data Unify representations by mapping between sources.

• Summary: Inference based on unified (general) knowledge representation.

Inference

• What is inference?

– Process of deriving new facts from old facts. “Reasoning”.

• What is a proof?

– A demonstration that a deductive inference is sound (= true, assuming logic is correct).

• What is logic?

– The rules used to perform deductive inference.

The Layer Cakeof the Semantic Web

OSM/OWL & Logic and Proof

• Model the application domain ontologically (high level of abstraction).

• Systematically transform application model into predicate calculus.

• Perform reasoning on predicates.

Ontology Language: OSM & OWL<owl:Ontology>…<owl:Class rdf:ID=“Italian”> <owl:unionOf rdf:parseType=“owl:collection”> <owl:Class ref:resource=“#Lazy”/> <owl:Class ref:resource=“#Mafioso”/> <owl:Class ref:resource=“#LatinLover”/> … <owl:Restriction> <owl:onProperty rdf:resource=“#hasName”/> <owl:minCardinality> <rdf:datatype=“&xsd:#nonNegativeInteger”>1</…> </owl:minCardinality> <owl:maxCardinality> …<owl:Class rdf:ID=“LatinLover”> <rdfs:subClassOf rdf:resource=“#Italian”/> <owl:disjointWith rdf:resource=“#Lazy”/> <owl:disjointWith rdf:resource=“#Mafioso”/>…<owl:ObjectProperty rdf:ID=“hasName”> <rdfs:domain rdf:resource=“#Italian”/> <rdfs:range rdf:resource=“#Name”/>...

Name Italian ItalianProf

Lazy LatinLoverMafioso

Lazy(x) => not ItalianProf(x)ItalianProf(x) => not Lazy(x)Mafioso(x) => not ItalianProf(x)ItalianProf(x) => not Mafioso(x)

1:*

has

1Name Italian ItalianProf

Lazy LatinLoverMafioso

Lazy(x) => not ItalianProf(x)ItalianProf(x) => not Lazy(x)Mafioso(x) => not ItalianProf(x)ItalianProf(x) => not Mafioso(x)

1:*

has

1

Transformation toPredicate Calculus: Example

Generated Predicates

1. Object Sets

2. Relationship Sets

Room(x), Room Nr(x), Cost(x), Date(x), Guest(x), Guest Nr(x),Current Guest(x), Future Guest(x), Guarantee Nr(x)

Room(x) has Room Nr(y),Room(x) has Cost(y),Guest(x) has reservation for Room(y) on Date(z),Guest(x) has Guest Nr(y)Future Guest(x) has Guarantee Nr(y)

3. Generated Rules

Referential-Integrity Constraintsxy(Room(x) has Room Nr(y) Room(x) Room Nr(y))...

Generalization/Specialization Constraintsx(Current Guest(x) Future Guest(x) Guest(x))

Participation Constraintsx(Room(x) 1y(Room(x) has Cost(y))x(Cost(x) 1y(Room(y) has Cost(x))...

Co-occurrence Constraints<x, y>(z(Guest(z) has reservation for Room(x) on Date(y)) 1w(Guest(w) has reservation for Room(x) on Date(y)))

A Valid InterpretationObject-Set Relations

Relationship-Set Relations

Constraints

RoomR1R2

Room Nr12

Cost9080

Room has Room NrR1 1R2 2

x(Room(x) 1y(Room(x) has Room Nr(y))...

...

...

Logic and Proof

Prove: ItalianProf(x) LatinLover(x)

Proof (by contradiction):1. LatinLover(x) negation of conclusion

2. ItalianProf(x) premise

3. ItalianProf(x) Italian(x) ontologically given

4. Italian(x) modus ponens (2&3)

5. Italian(x) Lazy(x) Mafioso(x) LatinLover(x) ontologically given

6. Lazy(x) Mafioso(x) LatinLover(x) modus ponens (4&5)

7. Lazy(x) Mafioso(x) resolution (1&6)

8. ItalianProf(x) Lazy(x) ontologically given

9. Lazy(x) modus ponens (2&8)

10. Mafioso(x) resolution (7&9)

11. Mafioso(x) ItalianProf(x) ontologically given

12. ItalianProf(x) modus ponens (10&11)

13. F contradiction (2&12)

Name Italian ItalianProf

Lazy LatinLoverMafioso

Lazy(x) => not ItalianProf(x)ItalianProf(x) => not Lazy(x)Mafioso(x) => not ItalianProf(x)ItalianProf(x) => not Mafioso(x)

1:*

has

1Name Italian ItalianProf

Lazy LatinLoverMafioso

Lazy(x) => not ItalianProf(x)ItalianProf(x) => not Lazy(x)Mafioso(x) => not ItalianProf(x)ItalianProf(x) => not Mafioso(x)

1:*

has

1

OWL Foundation: RDF (Resource Description Framework)

• A triple model:– Every assertion is decomposed in three parts

– (subject, predicate, object)

– For instance (tutorial.php, author, "Fabien")

• The subject is a URI identifying a resource.

• The predicate is a binary relation identified by a URI.

• The object is either a URI identifying a resource or a literal value.

• Each triple illustrated as a labeled arc.

• A set of statements/arcs is a graph of relations and attributes of URI resources.

RDFS (RDF Schema)

• A set of primitives to describe lightweight ontologies, allowing us to:

• Name the resource types and binary relation (property) types.

• Specify signature of properties:– Type of domain = type of subject

– Type of range = type of object

• Specify inheritance between classes (subClassOf);

• Specify inheritance between properties (subPropertyOf);

• Includes multiple inheritance.

RDF Graph

“Samuel /Baker/”

“M”

gc:name

gc:gender

#samuel

#sarah

#birthOfSamuelgc:born

gc:gaveBirth

“April 17, 1873”

“Chicago”

gc:date

gc:place

“Sarah /Baker/”

gc:name

#mark#marriagegc:married gc:married

“December 22, 1868”

“Boston”

gc:date

gc:place

gc:fathered

“Mark /Baker/”

gc:namerdf:type

gc:Marriage

rdf:typegc:Birth

rdf:type

gc:Individual

rdf:type

rdf:type

RDF vs. RDBMS/SQL

• Similarities:– Based on relational models

– Knowledge representations

– Integrity constraint / inference engines

• Differences:– Different relational models (e.g. only binary relations in RDF)

– In RDBMS, a missing tuple means that statement is false (closed-world assumption)

– In RDF, a missing tuple means unknown (open-world assumption)

• Links:– http://www.rdfabout.com/comparisons.xpd

N-ary Relation in RDF

• John buys a "Lenny the Lion" book from books.example.com for $15 as a birthday gift.

• There is a relation, in which individual John, entity books.example.com and the book Lenny_the_Lion participate.

• This relation has other components as well such as the purpose (birthday_gift) and the amount ($15).

N-ary Relation in RDF (Reification)

http://www.w3.org/TR/swbp-n-aryRelations/#example3

Querying RDF Data

• RDF data is a graph.– Resources/entities and attributes values are nodes.

– Relationships and properties are edges.

• Queries specify constraints on sub-graphs.

• Executing queries returns matching sub-graphs.

SPARQL

• Specifies queries over an RDF triple store

• Triple stores have an OWL/RDF schema

• Example: get names and, if available, gender and birthdate of people born in the 1870’s:

(An RDF Query Language)

SELECT ?Name ?Gender ?BirthDateWHERE { ?IndividualURI gc:name ?Name . OPTIONAL { ?IndividualURI gc:gender ?Gender ; gc:born ?Birth . ?Birth gc:date ?BirthDate } . FILTER REGEX(?BirthDate, "187\\d") }

Genesis

Querying RDF Data

• Links:– http://www.youtube.com/watch?v=3kdDKILwna0

– http://www.dailymotion.com/video/xewx1o_scribo-workbench-sparql-query-edito_tech

– http://www.youtube.com/watch?v=8F-gT02ZyDs