The Semantic Web: Current Status and Challenges

1
The Semantic WebCurrent Status and Challenges

• Stefan Decker
• Information Sciences Institute, University of
  Southern California
• Digital Enterprise Research Institute, National
  University of Ireland, Galway
• and collaborators
• Sergey Melnik, Wolfgang Neijdl, Bijan Parsia,
  Mario Schlosser, Michael Sintek

2
Ho! what have we here

• So very round and smooth and sharp? To me 'tis
  very clear This wonder of an Elephant Is very
  like a spear!
• John Godfrey Saxe (1816-1887)The Blind Men and
  the Elephant

3
Outline

• Semantic Web Overview
• Standards
• RDF
• Ontologies
• Research
• Rules
• Web Services Routing in P2P Networks
• Conclusion

4
Semantic Web

• coined by Tim Berners-Lee (1997)
• "The Semantic Web is an extension of the current
  web in which information is given well-defined
  meaning, better enabling computers and people to
  work in cooperation.
• T. Berners-Lee, J. Hendler, O. Lassila,The
  Semantic Web, Scientific American, May 2001

5
Motivation Why do we need the Semantic Web?

• Overabundance of Information
• Highly scattered and distributed
• Need to search and integrate information
• Cost for locating relevant information and
  deriving value from it is prohibitively
  expensive
• Reduce costs by
• Interconnecting workflows and business processes
• Enable data and service sharing
• Enable collaboration and information sharing
• Data and services sharing between diverse
  scientific groups such as genomic and biological
  sciences, and geosciences
• Across industrywide consortiums and standards
  bodies

6
How do we get there?
Research communities
DL, AI, DB, NLP, Networking
Standards bodies
W3C, OMG,
Non-profit
US, EC, (Japan?)
Industry
IBM, Nokia, HP, Microsoft(?),...
Business.semanticweb.org
7
Means to Achieve the Vision

• Explicit Ontologies and Interoperable Data
• Needed to understand each others data
• Web Services
• Required to actively interconnect
  systems(automatically make an appointment)

8
Technical challenges

• Interoperability
• Inaccurate, incomplete, heterogeneous data
• Unreliable, ill-defined, evolving services
• Natural language processing, data mining
• make information explicit
• Human-computer interaction
• querying interfaces, visualization
• Scalability
• Subsecond performance

9
Social challenges

• Standardization is hard
• DublinCore
• Bogus or inaccurate metadata
• Physician rating, profile
• Competition and commoditization
• Economical incentive
• Chicken and egg
• Complexity developers and users

10
Its the Economy, Stupid!

• PapiNet.org
• Vocabulary for Paper Industry
• BPMI.org
• Vocabulary for exchanging Business Process
  Models
• XML-HR
• Vocabularies for human resources (HR)
• DMTF Distributed Management Task Force
• Vocabularies for managing enterprises
• Wide range of E-Business standards available at
• http//www.oasis-open.org/home

11
Its the Knowledge, Stupid!

• Gen Ontology Working Group
• Annotations of Gene Sequences
• Earth Sciences (SCEC, GEON)
• Exchange of geology and earthquake data
• MeSH
• Annotations of medical research literature
  (MEDLINE)
• Bio-informatics
• Micro-array data markup language
• UMLS Metathesaurus
• Integration of Biomedical Vocabularies
• http//umlsinfo.nlm.nih.gov

12
Its the People, Stupid!

• E-commerce didnt invent the web, the web
  invented E-commerce
• Case Friend of a Friend (FOAF)
• Ontology driving community
• Trust-based models for
• Information and Semantics
• E-Commerce (E.g., E-bay)
• A little semantics, in the right place at the
  right time, can excite people
• But, the gain must be worth the pain
• Must have perceived end value
• The journey must seem not to horrible

13
Technology

• The Resource Description Framework
• Ontologies
• Rules
• Web Services/P2P Networks

14
Heterogenous Data

• To many data formats/languages

15
1. Step

• Define uniform, underlying syntax
• Lowest common denominator labeled
  graphs(semi-structured Data) -gt RDF

Relational Database
Structured Text (e.g., Vcard)
Person
begin vcardfn Stefann
DeckerStefanend vcard
Person
row
row
vcard1
fn
n
L-name
L-name
ID
ID
F-name
F-name
Stefan
DeckerStefan
1
Decker
Decker
Stefan
Birgit
2
16
XML

• Containment, hierarchy
• Adjacency (A followed by B)
• Attributes (atomic values)
• Opaque reference (IDREF)
• Good for serialization, poor for modeling
  relational semantics

17
Encoding of Information
The Creator of the Resource http//www.w3.org/Ho
me/Lassila is Ora Lassila
http//www.w3.org/Home/Lassila
Creator
Endless encoding possibilities in XML
18
Introduction to RDF

• RDF (Resource Description Framework)
• Beyond Machine readable to Machine understandable
• RDF unites a wide variety of stakeholders
• Digital librarians, content-raters, privacy
  advocates, B2B industries, AI...
• Significant (but less than XML) industrial
  momentum, lead by W3C
• RDF consists of two parts
• RDF Model (a set of triples)
• RDF Syntax (different XML serialization syntaxes)
• RDF Schema for definition of Vocabularies (simple
  Ontologies) for RDF (and in RDF)

19
A Simple Example

• Describing Resources
• URIs global OIDs, literals
• Binary relationships between objects
• Arcs (relationships) are first-class objects
• Blank (anonymous) nodes
• Ora Lassila is the creator of the resource
  http//www.w3.org/Home/Lassila
• Structure
• Resource (subject) http//www.w3.org/Home/Las
  sila
• Property (predicate) http//www.schema.org/Cre
  ator
• Value (object) "Ora Lassila

sCreator
http//www.w3.org/Home/Lassila
20
RDF

• Graph-based universal syntax

(Agent-) Applications
RDF-Layer (Single dataformat, Query and storage
System)
Scheduling Service
Insurance Ratings
Calendar
Semantics in a global, open environment?
21
Large scale Interoperation
Source
Destination
Likely to be implemented at the vocabulary
level!!!
NormalFault is_a Data consists of
----
Re-engineering
Translation Step
Abstraction
Adaptation
ltxsdschema xmlnsxsd"http//..."gt
ltxsdannotationgt A-Schema lt/xsd...lt/xsdsche
magt
DTD or XML Schema
Conceptual Domain Model(Objects and Relations)
22
Ontologies as KR Semantic Nets

• Semantic Web (and the Web) originally conceived
  as a (global, decentralized, etc.) Semantic Net
• Links determine meaning
• Link traversal significant
• Natural fit with hypermedia
• Two design traditions
• The Web Scruffies
• "Anything can say anything about anything"
• Principle of partial understanding
• Little semantics goes a long way
• Uniformity of representation and notation
• The Neats
• Coming out of the description logic tradition
• Decidability and practicality of key reasoning
  components key
• Formal semantics
• Economy and readability of notation

23
Step2 Ontologies

• What is an Ontology?
• An ontology is a specification of a
  conceptualization.
• Tom Gruber, 1993
• Ontologies are social contracts
• Agreed, explicit semantics
• Understandable to outsiders
• (Often) derived in a community process

24
Dynamic Communication Partners Interpretation
steps are too costly
25
Large scale Interoperation
Source
Destination
Likely to be implemented at the vocabulary
level!!!
NormalFault is_a Data consists of
----
Re-engineering
Translation Step
Abstraction
Adaptation
ltxsdschema xmlnsxsd"http//..."gt
ltxsdannotationgt A-Schema lt/xsd...lt/xsdsche
magt
DTD or XML Schema
Conceptual Domain Model(Objects and Relations)
26
OWL - Web Ontology Language

• OWL provides an RDF/XML vocabulary for defining
  classes, their properties and their relationships
  among classes.
• Based on Description Logics
• Enables to Classes and Properties
• OWL a W3C Candidate Recommendation (see
  http//www.w3.org/TR/owl-ref)

This part of the tutorial is a selection and
slight adaptation from an OWL tutorial from Roger
L. Costello and David B. Jacobs The MITRE
Corporation
27
Origins of OWL
DAML
OIL
RDF
All were influenced by RDF
DAML DARPA Agent Markup Language OIL Ontology
Inference Layer
DAMLOIL
OWL
28
OWL Full, OWL DL, and OWL Lite

• Not everyone will need all of the capabilities
  that OWL provides. Thus, there are three
  versions of OWL

OWL Full
OWL DL
OWL Lite
DL Description Logic
29
OWL Primitives for Defining Properties
Symmetric if P(x,y) then P(y,x) inverseOf if
P1(x,y) then P2(y,x) Transitive if P(x,y) and
P(y,z) then P(x,z) Functional if P(x,y) and
P(x,z) then yz InverseFunctional if P(x,y) and
P(z,y) then xz allValuesFrom P(x,y) has
yallValuesFrom(C) someValuesFrom P(x,y) has
ysomeValuesFrom(C) hasValue P(x,y) and
yhasValue(I) cardinality cardinality(P)
n minCardinality minCardinality(P)
n maxCardinality maxCardinality(P)
n equivalentProperty P1 P2
30
OWL Primitives for Defining Classes
intersectionOf C intersectionOf(C1, C2,
) unionOf C unionOf(C1, C2, ) complementOf
C complementOf(C1) oneOf C oneOf(I1, I2,
) equivalentClass C1 C2 disjointWith C1 !
C2 sameIndividualAs I1 I2 differentFrom I1 !
I2 AllDifferent I1 ! I2, I1 ! I3, I2 ! I3,
Thing C1, C2, , I1, I2, , P1, P2,
31
OWL ObjectProperty vs. DatatypeProperty
An ObjectProperty relates one Resource to another
Resource
ObjectProperty
Resource
Resource
A DatatypeProperty relates a Resource to a
Literal or an XML Schema datatype
DatatypeProperty
Resource
Value
32
Constructing Classes using Set Operators

• OWL gives you the ability to construct classes
  using these set operators
• intersectionOf
• unionOf
• complementOf

33
Defining a class using the intersectionOf
operator
Person
femalePerson
malePerson
Child
A father is a male Person with a least one Child.
Thus, a father may be defined as the
intersectionOf the malePerson class and an
anonymous class containing the hasChild property
with At least one value from Child..
34
Understanding intersectionOf
lt?xml version"1.0"?gt ltrdfRDF xmlnsrdf"http//w
ww.w3.org/1999/02/22-rdf-syntax-ns"
xmlnsrdfs"http//www.w3.org/2000/01/rdf-schem
a" xmlnsowl"http//www.w3.org/2
002/07/owl" xmlbase"http//www.
geodesy.org/water/naturally-occurring"gt
ltowlClass rdfIDFather"gt
ltowlintersectionOf rdfparseType"Collection"gt
ltowlClass rdfabout"malePerson"/gt
ltowlRestrictiongt
ltowlonProperty rdfresource"hasChildre
n"/gt ltowlallValuesFrom
rdfresource"Child"/gt ltowlmincardinality

rdfdatatype"http//www.w3.org/2001/XMLSchemanon
NegativeInteger"gt1
lt/owlmincardinalitygt
lt/owlRestrictiongt lt/owlintersectionOfgt
lt/owlClassgt ... lt/rdfRDFgt
This is read as A Father is the intersection of
the malePerson and an anonymous class that
contains a property hasChild and all values are
instances of Child. There is at least
one child." Here's an easier way to read this
A father is a male Person with at least one
child."
35
The cardinality is not mandating the number of
occurrences of a property in an instance document!

• Differentiate between
• 1. In an instance document there must be at least
  one child property for a father.
• 2. A father has at least one child.
• Difference
• 1. The first statement is an Integrity Constraint
• 2. The second statement is an assertion. It
  places no restrictions on the number of
  occurrences of the hasChildren property in an
  instance document. In fact, any Father resource
  may zero hasChildren properties. There must be
  one, however.
• Data validation under Description Logic Semantics
  not possible.

36
Assertions vs. ConstraintsExample Cardinality
Constraints
ltowlClass rdfIDFather"gt ltrdfssubClassOf
rdfresource"malePerson"/gt
ltrdfssubClassOfgt ltowlRestrictiongt
ltowlonProperty rdfresource"hasChil
dren"/gt ltowlmincardinality
rdfdatatype"http//www
.w3.org/2001/XMLSchemanonNegativeInteger"gt1lt/owl
mincardinalitygt lt/owlRestrictiongt
lt/rdfssubClassOfgt lt/owlClassgt
This is read as "The Father class is a
subClassOf malePerson, and a subClassOf an
anonymous class which has a property hasChildren.
There must at least one child for a father.
This is indicated by a cardinality of 1." Here's
an easier way to read this "The Father class is
a subClassOf malePerson. It has a property
hasChildren. There must be at least only one
child for a father."
37
Comparison
OWL Full OWL DL OWL Lite
All listed primitives. Further, you can mix RDF
Schema definitions with OWL definitions.
You cannot use owlcardinality with
TransitiveProperty. A DL ontology cannot
import an OWL Full ontology. You cannot use a
class as a member of another class, i.e., you
cannot have metaclasses. FunctionalProperty
and InverseFunctionalProperty cannot be used with
datatypes (they can only be used with
ObjectProperty).
All the DL restrictions plus You cannot use
owlminCardinality or owlmaxCardinality. The
only allowed values for owlcardinality is 0 and
1. Cannot use owlhasValue. Cannot use
owldisjointWith. Cannot use owloneOf. Cannot
use owlcomplementOf. Cannot use owlunionOf.
38
Advantages/Disadvantages

• Full
• The advantage of the Full version of OWL is that
  you get the full power of the OWL language.
• The disadvantage of the Full version of OWL is
  that it is difficult to build a Full tool. Also,
  the user of a Full-compliant tool may not get a
  quick and complete answer.
• DL/Lite
• The advantage of the DL or Lite version of OWL is
  that tools can be built more quickly and easily,
  and users can expect responses from such tools to
  come quicker and be more complete.
• The disadvantage of the DL or Lite version of OWL
  is that you don't have access to the full power
  of the language.

39
Ontology and Schema Languages A comparison
40
Ontology Editors Protégé-2000
ltrdfsClass rdfabout"mvMotorVehicle"gt
ltrdfssubClassOf rdfresource"rdfsResource"/gtlt
/rdfsClassgt ltrdfsClass rdfabout"mvPassengerV
ehicle"gt ltrdfssubClassOf rdfresource"mvMo
torVehicle"/gtlt/rdfsClassgt ltrdfProperty
rdfabout"mvrearSeatLegRoom"
amaxCardinality"1" arange"integer"gt
ltrdfsdomain rdfresource"mvMotorVehicle"/gt
ltrdfsrange rdfresource"rdfsLiteral"/gtlt/rdf
Propertygt
41
Selected Ontology Tools and Dimensions
42
Further Topics

• Ontology creation and learning
• Tools and Techniques for bootstrapping
• Tools and Techniques for enhancement of existing
  resources
• Mappings across ontologies and schemas
• Model Management and Ontology Algebras

43
The Layer Cake
Research Phase
Standardization Phase
Recommendation Phase

• Tim Berners-Lee
• Axioms, Architecture and Aspirations
• W3C all-working group plenary Meeting
• 28 February 2001

44
Rules
45
TRIPLE An RDF Query, Inference,and
Transformation Language
46
Motivation Why Rule Languages for the Web

• Declarative Processing
• Time to Market Faster to write rules than code
  for transformation, integration
• Rule capture part of the dynamic aspects of a
  domain
• Ontologies capture static aspects

47
Guiding Requirements for an RDF Rule Language

• Support RDF (graph-structured data)
• Handle multiple modeling semantics
• (OWL, DAMLOIL, UML, ER, TopicMaps, DAMLOIL,
  XML-Schema, Relational Data, special purpose data
  models)
• Special query systems for all of them?
• Distributed, heterogeneous sources
• Not all data is created equal

48
Basic Notion RDF Models

• Support for RDF models

Stefans Data
Franks Data
49
Implicit, Parameterized Models
50
TRIPLELanguage Overview

• Native support
• for Resources namespaces,
• Abbreviations
• Models (sets of RDF statements)
• Reification
• Rules with expressive bodies (full FOL syntax)
• Inspired by F-Logic
• subjectpredicate?object (molecule)
• Extended by Models, model expressions,
  parameterized models
• sp?o_at_m triple lts,p,ogt in model m
• sp?o_at_(m1 ? m2) model intersection, union, diff.
• sp?o_at_sf(m1, X, Y) Skolem function

51
Example Dublin Core

• dc http//purl.org/dc/elements/1.0/.
• isi http//www.isi.edu/.
• _at_isidocuments
• isid_01_01
• dctitle ? TRIPLE
• dccreator ? Stefan Decker
• dcsubject ? RDF
• dcsubject ? triples ... .

namespace abbreviations
model block
fact
52
Parameterized Models

• General format?P1, , Pn _at_model(P1, , Pn)
  clausesP1, , Pn
• Used for
• Data integration
• Model transformation
• Defining the semantics of languages layered on
  top of RDF (semantic spaces)
• Module system

53
Semantic Spaces Specifying Semantics via
Parameterized Models

• RDF Schema, UML (and other frame/OO
  systems)semantics can be directly defined in
  TRIPLE as a parameterized model
• OIL, DAMLOIL, OWL (i.e., expressive ontology
  languages, DL)requires interaction with foreign
  reasoning components (e.g., DL classifier)

model
sem(model)
? Mdl _at_sem(Mdl) clauses
rules describing the semantics of a data model
54
Specification of RDF Schema Semantics
namespace abbreviations

• rdf 'http//www.w3.org/...rdf-syntax-ns'.
• rdfs 'http//www.w3.org/.../PR-rdf-schema-...'
  .
• type rdftype.
• subPropertyOf rdfssubPropertyOf.
• subClassOf rdfssubClassOf.
• FORALL Mdl _at_rdfschema(Mdl)
• FORALL O,P,V OP-gtV lt-
• OP-gtV_at_Mdl.
• FORALL O,V OsubClassOf-gtV lt-
• EXISTS W (OsubClassOf-gtW
• AND WsubClassOf-gtV).

resource abbreviations
model block
copy triples from Mdl
Transitivity of subClassOf
55
Example Cars Ontology

• _at_cars
• xyzMotorVehiclerdfssubClassOf -gt
  rdfsResource.
• xyzPassengerVehiclerdfssubClassOf -gt
  xyzMotorVehicle.
• xyzTruckrdfssubClassOf -gt
  xyzMotorVehicle.
• xyzVanrdfssubClassOf -gt xyzMotorVehicle.
• xyzMiniVan
• rdfssubClassOf -gt xyzVan
• rdfssubClassOf -gt xyzPassengerVehicle.

xyzMotorVehicle
xyzTruck
xyzVan
xyzPassengerVehicle
xyzMiniVan
X xyzVan X xyzTruck X
xyzPassengerVehicle
FORALL X lt- XrdfssubClassOf -gt
xyzMotorVehicle_at_cars.
X xyzVan X xyzTruck X
xyzPassengerVehicle X xyzMiniVan
FORALL X lt- XrdfssubClassOf -gt
xyzMotorVehicle_at_rdfschema(cars).
56
Specification of UML Semantics
rdf 'http//www.w3.org/...rdf-syntax-ns'. uml
'http//www.omg.org/uml/1.3/'. FORALL Mdl
_at_uml(Mdl) FORALL O,P,V OP-gtV lt-
OP-gtV_at_Mdl. FORALL X,Z g(X,Z)rdftype-gtuml
Generalization
uml'Generalization.child'-gtX
uml'Generalization.parent'-gtZlt- EXISTS
Y,G1,G2 G1uml'Generalization.child'-gtX
uml'Generalization.parent'-gtY AND
G2uml'Generalization.child'-gtY
uml'Generalization.parent'-gtZ .
Transitivity of Generalization
57
DAMLOIL Semantics

• daml_oil(Mdl) model realized by accessing a DL
  classifier (e.g., Racer or FaCT)
• access only allowed in rule bodies
• realization Mdl is materialized and transformed
  into input for DL classifier classifier is
  invoked (direct) subClassOf and sameClassAs
  added to daml_oil(Mdl) model rest handled via
  TRIPLE rules directly

rules for remaining semantics
ontology
daml_oil(ontology)
subClassOfsameClassAs
DL classifier
mat.
?O _at_daml_oil(O) clauses

• results in hybrid rule language similar to Carin,
  but more pragmatic approach powerful but
  incomplete

58
DAMLOIL Example

• daml 'http//www.daml.org/.../damloil'.
• animals 'http//www.example.org/animals'.
• _at_animalsontology
• animalsAnimalrdftype -gt damlClass.
• animalsHerbivorerdftype -gt damlClass
• damlsubClassOf -gt animalsAnimal.
• animalsCarnivorerdftype -gt damlClass
• damlsubClassOf -gt animalsAnimal
• damldisjointWith -gt animalsHerbivore.
• animalsOmnivorerdftype -gt damlClass
• damlsubClassOf -gt animalsHerbivore
• damlsubClassOf -gt animalsCarnivore.
• FORALL C
• lt- CdamlsameClassAs -gt damlNothing_at_daml_oil(
  animalsontology).

Animal
s
s
Herbivore
Carnivore
damldisjointWith
s
s
Omnivore
s damlsubClassOf
find all unsatisfiable classes(will detect
Omnivore)
59
Realization Mapping to Horn Logic

• First implementation by mapping to Horn Logic /
  XSB system (Prolog with tabled resolution)
• model theory for full logic completed
• Lloyd-Topor transformation for quantifiers etc.
• RDF-specific transformations given as rewrite
  rules

60
Realization Compilation to Horn logic

• First implementation (and informal semantics) by
  mapping to Horn Logic / XSB system (Prolog with
  tabled resolution)
• Lloyd-Topor transformation for quantifiers etc.
• RDF-specific transformations given as rewrite
  rules

61
triple.semanticweb.org
62
Web Services
63
Web Services vs. the (Semantic) Web

• Semantic Web To do for KR what the Web did for
  hypertext
• Web services To exploit Web infrastructure for
  distributed applications
• The problem of Discovery
• Suppose the set of WSs grows like the Web
• Difficulties
• Finding compatible services
• Finding services with desired functionality
• Finding services with desired other properties
  (cost, QOS, location)
• Controlling effort put into these searches
• All this is much more difficult for compositions
  (or possible compositions)

Slides from Bian Parsia
64
UDDI and the Problem of Discovery

• UDDI is "a 'meta service' for locating web
  services by enabling robust queries against rich
  metadata." --UDDI 3.0 Specification
• UDDI is a Web service
• It has an API (on the Web).
• Easy to conceptualize as an application
• UDDI has a "rich" metadata model
• Developed specifically for UDDI
• "Until now, there has been no central way to
  easily get information about what standards
  different services support and no single point of
  access to all markets of opportunity, allowing
  them to easily connect with all potential service
  consumers." (pg. 1)
• Until now, the Web has shown that central ways,
  single points of access, aiming for all
  markets just dont work at Web scale
• Good bet From now, the Web will continue to show
  this
• The UDDI players (e.g., Microsoft) tried this
  before and lost

Slides from Bian Parsia
65
Fixing UDDI

• Evolutionary (Focus on simple Discovery)
• Why should WSs reinvent the wheel?
• Treat Semantic Web tech and standards like
  current Web tech and standards
• Natural progression
• First keywords, then tModels
• Then "taxonomies
• Then incorporating Semantic Web ontologies
• Mapping DAML-S profiles into tModels
• Then moving from tModels to Ontologies
• Radical
• Decentralize, decentralize, decentralize
• Design representations for automatable reasoning
• And be expansive about what sorts of reasoning
  you desire
• Small gains add up!
• Doing it right, or well, or both, is worth it

Slides from Bian Parsia
66
Ontology-based Service Discovery in P2P-Networks
67
Semantic Routing (for Data, Queries)

• Route information based on content and ontologies
  instead of IP-addresses
• Applications
• Discovery of distributed Web Services in
  Peer-to-Peer Networks
• Motivation Avoid centralized database (UDDI -
  single point of failure, man in the middle,
  freshness)
• Publish/Subscriber Models
• Metadata-based File Exchange in P2P networks

68
P2P Infrastructure for Semantic Web Services

• (Semantic) Web Services
• Large network of service providers capable of
  instantiating high-level task descriptions in
  distributed fashion
• How to reach all service providers that are
  potentially interesting?

You Are Here
Service Providers
69
Idea

• Define a multi-dimensional Overlay Network (on
  top of the IP-network) for a P2P network
• Map Ontology-terms to network dimensions
• Each dimension identifies a term
• Route information to dimension
• Cayley-graphs provide a good starting point
• Tricky part
• Keep network organized when nodes join and leave
  the network
• Define Mapping

70
Cayley Graphs

• Graph representing a permutation group G,
  described by a set of generators Akers,
  Krishnamurty
• Regular, vertex-symmetric, recursively
  decomposable
• Optimal routing and broadcast algorithms exist

111
000
Hypercube
Star Graph
71
Hypercube Topology

• Broadcast algorithm
• Tag message with dimension of link on which it is
  sent and forward message only on links of higher
  dimension
• Properties
• Network diameter, characteristic path length,
  node degree are of O(logbN)
• Fault tolerance, vertex symmetry

Step 2
Step 3
Step 1
72
Topology Construction Algorithm Sketch

• Topology of next biggest complete hypercube is
  always implicitly present in any current network
  topology
• Allows for hypercube algorithms (broadcast,
  search) to run
• Node departures Neighbors of a departing node
  jump in to cover the position(s) previously
  occupied and covered by the departing node
• Complete hypercube topology is collapsed and
  stored among the existing nodes, allowing for any
  number of nodes in the network
• Node arrivals Collapsed topology is
  reconstructed, new node takes over responsibility
  for one or more positions
• Unfold topology by retrieving topology
  information from nodes in the network

73
Topology Construction I
II
I
74
Topology Construction II
IV
III
75
Topology Construction III
VI
V
76
Topology Construction IV
VI
V
3-Hypercube
77
Ontology based Routing

• Goal Use additional global knowledge to improve
  search performance of P2P network
• Contain broadcast of search messages to
  potentially interesting peers
• Approach Partitioning of network into concept
  clusters
• Clusters are assigned to concepts organized in an
  ontology

Service Ontology
Domain Ontology
78
HyperCuP Network Construction I

• Use concepts A, D, E, F as structuring concepts
  C0..C3

79
HyperCuP Network Construction II

• Create concept cluster Cayley graph by grouping
  similar peers

Peer address Storage coordinates 3214
80
HyperCuP Network Construction III

• Link concept clusters by outer hypercube
  topology

ØDelivery Ù ØRetail Ù Wholesale Ù Motor Vehicles
Delivery Ù ØRetail Ù ØWholesale Ù Motor Vehicles
Peer address Storage coordinates 3214
Concept coordinates 1001
81
Querying the HyperCuP Network I

• Queries Logical conjunctions and disjunctions of
  negated and non-negated ontology concepts
• Example Wholesale Ú (Delivery Ù Retail) C2 Ú
  (C0 Ù C1)
• Logical minimization of query
• To retrieve logical minterms in query
• Minterm locality
• Minterms represent larger clusters of concept
  clusters

C0C1C2C3
82
Querying the HyperCuP Network II

• Routing to concept clusters
• Broadcast search message among concept clusters
  as determined by minterms
• HyperCuP Broadcast in concept clusters
• Inform all peers inside all addressed clusters

1.
4.
4.
3.
0.
3.
2.
4.
4.
Wholesale Ú (Delivery Ù Retail) C2 Ú (C0 Ù C1)
3.
83
p2p.semanticweb.org
84
Outlook From the Semantic Web to the Semantic
Desktop

• Who is satisfied with his/her current Computer
  Desktop?
• Co-evolution enabled between the Semantic Web and
  the Semantic Desktop
• Metadata creation and Application integration

85
Conclusion

• The Semantic Web is forced by various
  developments (e-Commerce, e-Science, e-Society)
• Addition of Semantics add new perspective on
  Rules, Distributed Computing, many other things
  in Computer Science

86
Conclusion II

• The best way to predict the future is to invent
  it
• (be part of it you make a difference!)

87
Collaborators

• This presentation would not have been possible
  without the help, hard work, enthusiasm, and
  slides of many people I was lucky enough to work
  with
• Sergey Melnik
• Michael Sintek
• Mario Schlosser
• Martin Lacher
• Wolfgang Nejdl
• Yuhui Jin
• Prasenjit Mitra
• Gio Wiederhold
• Bijan Pasia