What`s Wrong with the Semantic Web?

What‘s Wrong with the Semantic Web
XML-Days
Berlin, 25. September 2006
Name und Datum
www.uni-mannheim.de
Seite 1
This talk is NOT about
•
•
•
•
Name und Datum
General OWL-bashing
Why databases can solve everything anyways
Web 2.0, Tagging, Social-semantic software, …
Semantic Desktop, semantic Grid, semantic web
services, semantic …..
www.uni-mannheim.de
Seite 2
This Talk IS about
•
•
•
•
•
•
Name und Datum
The Vision of the Semantic Web
Semantic Web Technologies in 10 Minutes
The typical Semantic Web Application in 2006
Problems: Where the assumptions break
A (personal) perspective on solutions
The typical semantic web application in 2010
(hopefully)
www.uni-mannheim.de
Seite 3
The Original Vision
• Berners-Lee, Hendler,
Lassila: The Semantic Web,
Scientific American, May
17th 2001
• Cited (and abused)
extensively in literature and
science marketing
Name und Datum
www.uni-mannheim.de
Seite 4
Key Ideas
•
•
•
•
Name und Datum
Smart Devices
Personal Information Agents
Knowledge about objects, time and space
Trusted Information
www.uni-mannheim.de
Seite 5
Observations
• Part of the traditional Web
• Paradoxes, incompleteness and other logical fallacies
• Ontology Integration and Ontology Mappings
• Evolution of Knowledge and Meaning
Name und Datum
www.uni-mannheim.de
Seite 6
Key Technologies
• Machine-Readable Metadata
– Based on XML and RDF
• Logic, Inference Rules and Proofs
• Ontologies
• Agent Technologies
(nowadays read: „web services“)
Name und Datum
www.uni-mannheim.de
Seite 7
RDF: Resource Description Framework
•
RDF is a data model
– used to describe meta-data of a piece of data
– not a language, like XML
• although it has an XML syntax (but also other)
•
Benefits:
– Unique representations of content objects
– Explicit relations between resources
last-name
http://ki.informatik.
uni-mannheim.de/people/#heiner
photo
“Stuckenschmidt”
gives-lecture
http://ki.informatik.
uni-mannheim.de/courses/#sw
Name und Datum
title
www.uni-mannheim.de
“Semantic Web
Technologies”
Seite 8
(X R Y), (R subPropertyOf Q) Æ (X Q Y)
RDF Schema
(X R Y), (R domain C) Æ (X type C)
(X type C), (C subClassOf D) Æ (X type D)
Person
subClassOf
domain
Teacher
range
teaches
Course
subPropertyOf
type
type
gives-lecture
RDF schema
RDF data
teaches
http://ki.informatik.
uni-mannheim.de/people/#heiner
Name und Datum
gives-lecture
www.uni-mannheim.de
http://ki.informatik.
uni-mannheim.de/courses/#sw
Seite 9
Problem: no semantic guarantees
Source B
Does_consultancy_for
domain
range
type
www.bigcompany.com
Does_consultancy_for
http://ki.informatik.
uni-mannheim.de/people/#heiner
Name und Datum
Company
type
gives-lecture
www.uni-mannheim.de
http://ki.informatik.
uni-mannheim.de/courses/#sw
Source A
Seite 10
Logical Reasoning about Resources
• Logical Axioms limit
allowed interpretations:
Ontology
Person
Thing
Course
Company
Teacher ∧ Person ∧ Thing ∧ ¬Company ∧ Company
Teacher
Teacher ∧ Person ∧ Thing ∧ ¬Company
Teacher ⇔ ∃teaches.Course
Teacher ∧ Person
Teacher
Teacher ⇒ Person
Company
Person ⇒ Thing ∧ ¬Company
type
http://ki.informatik.
uni-mannheim.de/people/#heiner
Name und Datum
teaches
www.uni-mannheim.de
http://ki.informatik.
uni-mannheim.de/courses/#sw
Seite 11
The Web Ontology Language OWL
•
Full
OWL Lite:
ƒ Classification hierarchy
ƒ Simple constraints
•
DL
OWL DL:
ƒ Maximal expressiveness
ƒ While maintaining tractability
ƒ Standard formalisation
•
Lite
RDF
Schema
OWL Full:
ƒ
ƒ
ƒ
ƒ
Name und Datum
Very high expressiveness
Loosing tractability
Non-standard formalisation
All syntactic freedom of RDF
(self-modifying)
Syntactic
Syntacticlayering
layering
Semantic
Semanticlayering
layering
www.uni-mannheim.de
Seite 12
Where are we now
ontology
edit
extract
instances
Name und Datum
www.uni-mannheim.de
Seite 13
Living in the Real Web
1. The Technical Level
–
–
–
Distributed Information
P2P-like Architecture
Possible Failures
2. The Content Level
–
–
–
Name und Datum
Inconsistency and Incompleteness
Heterogeneity and Ontology Alignment
Multimedia Information Extraction
www.uni-mannheim.de
Seite 14
Where it breaks
ontology
edit
generate
import
instances
Name und Datum
www.uni-mannheim.de
Seite 15
Modular Ontologies
Name und Datum
www.uni-mannheim.de
Seite 16
The Case for Modularization
•
Distributed Development, Maintenance and use
– Experts can update their portion independently of other parts
•
Selective Publication and Use of Terminologies
– Stable subsets can be published in the development phase
– Users can chose relevant subset of an ontology
•
Manual Inspection and Validation
– Small, coherent modules are easier to understand
•
Editing, Visualization and Reasoning
– Available tools do not scale to very large ontologies
Name und Datum
www.uni-mannheim.de
Seite 17
Partitioning Ontologies
• Scenario 1: Structured System
• Idea: create a network where each node is in charge of one
partition and distribute data and metadata accordingly
Advantages:
• Reduction of bottlenecks
• No single point of failure
• Possibiliy for semanic routing
based on ontology structure
Name und Datum
www.uni-mannheim.de
Seite 18
Partitioning Ontologies
• Scenario 2: Unstructured System
• Idea: Distribute Partitions in an optimal way in an existing
network (ontology should be close to the point of use)
Advantages:
• local processing
• Use-based partitioning
• Improvement of communication
based on concept similarity
Name und Datum
www.uni-mannheim.de
Seite 19
So your ontologies are distributed…
• current reasoners:
– Global ontology
– Reason in global
ontology
• Problems:
–
–
–
–
Name und Datum
Scalability?
Reasoning specificity?
Privacy? Autonomy?
Robustnes?
www.uni-mannheim.de
Seite 20
…your Reasoning should be as well
• Alternative approach:
– Local reasoning
– Suitable combination
• Requirements:
– Formal framework
– Reasoning algorithm
– The reason-able
system implementation
Name und Datum
www.uni-mannheim.de
Seite 21
A Formal framework: C-OWL in a nutshell
• Distributed Description Logics (DDL)
– Captures the case of multiple ontologies pairwise linked by
semantic mappings
• Ontologies:
– DL knowledge bases
(T-box and A-box)
T-box
A-box
M-box
• Mappings:
– Bridge rules
(M-box)
Name und Datum
M-box
T-box
A-box
M-box
M-box
T-box
A-box
www.uni-mannheim.de
T-box
A-box
Seite 22
Reasoning
in C-OWL
Galaxy
Neptune: B
A
• Similarly to OWL-DL:
– Core reasoning task in DDL –concept A
B
subsumption
• Difference from OWL-DL:
– Scope – Galaxy
– Mappings matter
• Subsumption in DDL –
a global subsumption
Name und Datum
www.uni-mannheim.de
Earth
D
C
D C
Seite 23
Distributed reasoning algorithm
Cycle is detected
Local
reasoner
Propagation
Propagation
Propagation
Local
reasoner
Local
reasoner
Propagation
Reasoning
is finished
Local
reasoner
Local
reasoner
Name und Datum
Propagation
www.uni-mannheim.de
Seite 24
DRAGO
Distributed Reasoning Architecture for Galaxy of Ontologies
DRAGO Manager 1
Name und Datum
www.uni-mannheim.de
DRAGO Manager 2
DRAGO Manager 3Seite 25
Where do the mappings come from ?
• Decompositional Approach:
– One existing Ontology is split into different ones
– Mappings arise naturally from decomposition
• Compositional Approach:
– Different Ontologies exist
– Mappings have to be found based on semantic
correspondences
Name und Datum
www.uni-mannheim.de
Seite 26
Heterogeneous Classifications
SEA
IMAGES
SEA
EVENING
MOUNTAIN
HOTELS
CHEAP EXPENSIVE
MORNING
‘IMAGES OF THE SEA IN THE MORNING’
Name und Datum
PICTURES
LAKE
VERY
EXPENSIVE
‘PICTURES OF THE SEA’
‘IMAGES OF THE SEA IN THEwww.uni-mannheim.de
MORNING’
⊆
‘PICTURES OF THE SEA’
Seite 27
Ontology Matching
Name und Datum
www.uni-mannheim.de
Seite 28
Result: Mappings
• Mapping elements are 5-Tuples: (id,e,e‘,R,n)
–
–
–
–
id is a unique identifier for a given mapping element
e and e‘ are entities in the mapped ontologies
R is a relation that holds between the elements
n is a confidence measure for the mapping
• Two possible readings the measure
– Degree to which the entities relate , but also
– Confidence that the result of the matching is correct
Name und Datum
www.uni-mannheim.de
Seite 29
Problem: Partial Matching
• Sloppy terminologies need robust inference
almost subClassOf
Name und Datum
www.uni-mannheim.de
Seite 30
Reasoning with probabilistic Mappings
• Ontology on Peer 1: (O1)
Name und Datum
• Ontology on Peer 2: (O2)
www.uni-mannheim.de
Seite 31
Normalization: Ontologies as Rules
Company v ∀isLocatedIn.City
Company v ∀hasEmployee.Person
Company v ∀hasName.String
Person v ∀hasName.String
City v ∀hasName.String
Company(X), isLocatedIn(X,Y) Æ City(Y).
Company(X), hasEmployee(X,Y) Æ Person(Y).
Company(X), hasName(X, Y) Æ String(Y).
Person(X), hasName(X,Y) Æ String(Y).
City(X), hasName(X,Y) Æ String(Y).
Name und Datum
www.uni-mannheim.de
Seite 32
Bayesian Logic Programs
man(Bob).
married(Bob, Alice).
worksFor(Bob, Microsoft).
worksFor(Alice, Microsoft).
woman(Y) | man(X), married(X, Y).
Name und Datum
www.uni-mannheim.de
Seite 33
Bayesian Logic Programs
(example cont’d)
enterprise(Y) | man(X), worksFor(X, Y).
enterprise(Y) | woman(X), worksFor(X, Y).
Name und Datum
www.uni-mannheim.de
Seite 34
Translation to Bayesian Network
(CPTs omitted, see previous slide)
Name und Datum
www.uni-mannheim.de
Seite 35
Problem: Wrong Mappings
• Example 1: Matching Fallacies
–
–
–
–
i:Author ⊆ Person
j:Authorization ⊆ ¬Person
i:Person ≡
j:Person
≡
i:Author
j:Authorization
• Example 2: Modelling Fallacies
–
–
–
–
Name und Datum
i: SportsCar ⊆ Car
j:UselessThings ≡ ¬UselessThings
⊆
i:Car
j: UsefulThing
⊆
i:SportsCar
j: ¬UsefulThing
www.uni-mannheim.de
Seite 36
Reasoning ABOUT Mappings
• Check formal properties of mappings:
– Are there inconsistent mappings
– Are there redundant mappings
– Are there implied mappings
• Use Cases:
– Support for manual mapping creation
– Validation of automatically created mappings
Name und Datum
www.uni-mannheim.de
Seite 37
Subsumption propagation in DDL
DTB = 〈T1, T2, B12〉
T1
T2
A
G
isA
DTB
isA
H
B
GI2 ⊆ r12(AI1) ⊆ r12(BI1) ⊆ HI2
Name und Datum
www.uni-mannheim.de
Seite 38
Example 1: Inconsistency
AuthorizationI2 = ∅
T1
T2
Authorization
⊆ ¬Person
Author
isA
DTB
isA
Person
Person
AuthorizationI2 ⊆
r12(AuthorizationI1) ⊆ r12(PersonI1) ⊆ PersonI2
Name und Datum
www.uni-mannheim.de
Seite 39
Example 2: Embedding
r12(SportsCar) = ∅
T1
T2
SportsCar
UselessThings
isA
UseFullThings
Car
r12(SportsCarI1) ⊆ UselessThingsI2
r12(SportsCarI1) ⊆ r12(CarI1) ⊆ UseFullThingsI2
Name und Datum
www.uni-mannheim.de
Seite 40
Repairing Mappings
T1
Author
T2
≡
(n = 0.47)
(n = 1.0)
isA
Authorization
⊆ ¬Person
DTB
isA
(n = 0.47)
Person
1.
2.
3.
Name und Datum
≡
(n = 1.0)
Person
Syntactic Matching
Structural Matching
Analysis
www.uni-mannheim.de
Seite 41
Repairing Mappings
T1
≡
T2
(n = 0.47)
Author
(n = 1.0)
isA
Authorization
⊆ ¬Person
DTB
isA
(n = 0.47)
Person
1.
2.
3.
Name und Datum
Syntactic Matching
Structural Matching
Analysis
≡
4.
5.
(n = 1.0)
Person
Compute Conflict Sets
Select Problematic Rules and
repair mapping
www.uni-mannheim.de
Seite 42
The typical Web Application 2010 (?)
Name und Datum
www.uni-mannheim.de
Seite 43
Conclusions
• So what is wrong with the semantic web so far ?
– A lot of work was done on language for describing rich
information semantics (which is good!)
– Too little attention has been paid to the specific needs of a
distributed environment such as the Web (this is not so good)
• Is it any good then ?
– YES. There are many useful applications with a rather
centralized nature (community portals, company web sites)
– YES. People start recognizing the need for distributed abd
robust approaches.
Name und Datum
www.uni-mannheim.de
Seite 44
The KR & KM Research Group
• Christian Meilicke
• Livia Predoiu
• Anne Schlicht
• Heiner Stuckenschmidt
Name und Datum
www.uni-mannheim.de
Seite 45
Acknowledgement
•
Thanks to Rainer Eckstein for the invitation
•
During the Presentation, I used material originally created by:
–
–
–
–
Luciano Serafini, Andrei Tamilin, ITC-IRST Trento
Frank van Harmelen, Vrije Universiteit Amsterdam
Jerome Euzenat, INRIA Rhones-Alpes
Pavel Shvaiko, Fausto Giunchiglia, University of Trento
•
The DRAGO System for distributed reasoning with ontologies can be
downloaded at: http://sra.itc.it/projects/drago/
•
The work is partially funded by the German Science Foundation in the EmmyNoether Programme
Name und Datum
www.uni-mannheim.de
Seite 46