forschungszentrum informatik, karlsruhe fzi research center for information science at the...

Forschungszentrum Informatik, Karlsruhe

FZI FZI Research Center for Information ScienceResearch Center for Information Scienceat the University of Karlsruheat the University of Karlsruhe

Variance in e-Business

Service Discovery

Stephan Grimm,

Boris Motik,

Chris Preist (HP Labs, Bristol)

Slide 2

OverviewOverview

• Introduction

• Intuition behind modelling service semantics

• Operationalising discovery using logic

• Matching service descriptions

• Conclusion

Slide 3

Service Discovery in the Semantic WebService Discovery in the Semantic Web

• Service– Web Service vs. high-level eBusiness Service

• Service Discovery– Locating Providers who meet a Requestor´s needs

– Based on Semantic Descriptions of Services

• Semantic Description of a Service– Describing the Capabilities of the Service

– Using ontology languages, such as OWL

– Referring to common domain ontologies

Slide 4

OverviewOverview

• Introduction

• Intuition behind modelling service semantics

• Operationalising discovery using logic

• Matching service descriptions

• Conclusion

Slide 5

Service Description – Service InstanceService Description – Service Instance

shipping1

BremenPlymouth

from to

packageXitem

50 kg

weight

set of accepted Service Instances

. . .

shipping2

HamburgDover

from to

barrelYitem

25 kg

weight

Service Instances

Service Description

Shipping containers from UK to Germany describes

Slide 6

Variance in Service DescriptionsVariance in Service Descriptions

• Two kinds of variance in service descriptions

toshipping2 Hamburg

. . .

toshipping1 Bremen

toshipping3 Boston

– due to incomplete knowledge

. . .

toshipping2 Hamburg

toshipping1 Bremen

Shipping to Germany

– due to intended diversity

differentservice

instances . . . differentpossible worlds

Slide 7

Discovery by Matching Service DescriptionsDiscovery by Matching Service Descriptions

• Matching Service Descriptions of Requestors an Providers

• If there are common instances, requestor and provider can (potentially) do business with each other

(Sr)I  ∩  (Spi)I ≠ Ø

Sr

ServiceRequestor

Spi

ServiceProviders

Sp1

Spn

...

– How do their Service Descriptions intersect ?

Slide 8

OverviewOverview

• Introduction

• Intuition behind modelling service semantics

• Operationalising discovery using logic

• Matching service descriptions

• Conclusion

Slide 9

Intuition Intuition ↦↦ DL DL

• Service Description ↦ set of DL axioms D={1, ... , n}

– A service concept S occurring in some i

• Domain Knowledge ↦ DL knowledge base KB

Slide 10

Intuition Intuition ↦↦ DL DL

• Possible World ↦ Model I of KB ∪ D

• Service Instance ↦ relational structure in I

• acceptable Service Instances ↦ Extension SI of S

• Variance due to intended diversity ↦ |SI| ≥ 1

• Variance due to incompl. knowl. ↦ several Models I1, I2, ...

• Matching ↦ boolean function match(KB, Dr, Dp)– way of applying DL inferences

(Sr)I1

(UKCity)I1

(City)I1

item

fromfrom

(Package)I1 (Sr)I2

(UKCity)I2

(City)I2

item

from

(Package)I2

. . .

Slide 11

Towards Intuitive Modelling PrimitivesTowards Intuitive Modelling Primitives

Characterising Property Restrictions

• Multiplicity– single-valued

– multi-valued

• Variety– fixed value

– value range

• Availability– Mandatory

– obligatory

• Range Coverage– Covering

– non-covering

Slide 12

OverviewOverview

• Introduction

• Intuition behind modelling service semantics

• Operationalising discovery using logic

• Matching service descriptions

• Conclusion

Slide 13

Treating Variance in MatchingTreating Variance in Matching

• Resolving Incomplete Knowledge holds in every possible world :

Entailment KB ∪ Dr ∪ Dp ⊨ holds in some possible world :

Satisfiability KB ∪ Dr ∪ Dp ∪ {} sat.

• Resolving Intended Diversity– Request and Capability overlap :

Non-Disjointness = Sr ⊓ Sp ⋢ ⊥

– Request more specific than Capability :Subsumption = Sr ⊑ Sp

– Capability more specific than Request :Subsumption = Sp ⊑ Sr

⊨ sat.

⊑ ⊓

Slide 14

DL Inference for MatchingDL Inference for Matching

• Satisfiability of Concept Conjunction

(Sr ⊓ Sp) is satisfiable w.r.t. KB ∪ Dr ∪ Dp

(Sr)I1

(Sp)I1

. . .(Sr)I2

(Sp)I2

⊨ sat.

⊑ ⊓X

• (Sr)I ∩ (Sp)I ≠ Ø in some possible world

• Intuitiuon:– incomplete knowledge issues can be resolved such that request and

capability overlap

Slide 15

Satisfiability of Concept ConjunctionSatisfiability of Concept Conjunction

• Example:

⊨ sat.

⊑ ⊓X

• match(KB, Dr, DpA) = true

• match(KB, Dr, DpB) = true

– UKCity ⊓ USCity ⊑ ⊥ is not specified in KB

(Sr)I(UKCity)I

(City)I

Plymouth

Dublin from

from

(SpA)I

(USCity)I

(SpB)I

(Sr ⊓ Sp) is satisfiable

w.r.t. KB ∪ Dr ∪ Dp

Slide 16

DL Inference for MatchingDL Inference for Matching

• Entailment of concept subsumption

KB ∪ Dr ∪ Dp ⊨ Sr ⊑ Sp

• (Sr)I (Sp)I in every possible world

• Intuition:– the request is more specific than the capability regardless of how

incomplete knowledge issues are resolved

⊨ sat.

⊑ ⊓X

(Sr)I1

(Sp)I1

. . .

(Sr)I2

(Sp)I2

Slide 17

Entailment of Concept SubsumptionEntailment of Concept Subsumption

• Example:

• match(KB, Dr, DpA) = false

– Dublin outside the UK

⊨ sat.

⊑ ⊓X

(Sr)I

(UKCity)I

(City)I

Plymouth

Dublin from

from

(SpA)I

KB ∪ Dr ∪ Dp

⊨ Sr ⊑ Sp

Slide 18

DL Inference for MatchingDL Inference for Matching

• Entailment of Concept Non-Disjointness

KB ∪ Dr ∪ Dp ⊨ Sr ⊓ Sp ⋢ ⊥

• (Sr)I ∩ (Sp)I ≠ Ø in every possible world

• Intuition:– the request and the capability overlap regardless of how incomplete

knowledge issues are resolved

(Sr)I1

(Sp)I1

. . .

(Sr)I2

(Sp)I2

⊨ sat.

⊑ ⊓X

Slide 19

Entailment of Concept Non-DisjointnessEntailment of Concept Non-Disjointness

• Example:

• match(KB, Dr, DpA) = true

• match(KB, Dr, DpA) = false

– Plymouth outside the US in at least one possible world

⊨ sat.

⊑ ⊓X

(Sr)I(UKCity)I

(City)I

Plymouth

Dublin from

from

(SpA)I

(USCity)I

(SpB)I

KB ∪ Dr ∪ Dp

⊨ Sr ⊓ Sp ⋢ ⊥

Slide 20

Practicability of InferencesPracticability of Inferences

• Satisfiability of Concept Conjunction– very weak : vulnerable to false positive matches

– relies on additional disjointness constraints in domain ontologies

• Entailment of Concept Subsumption– Very strong : misses intuitively correct matches

• Entailment of Concept Non-Disjointness– Tries to overcome deficiencies of the other two inferences

– relies on range-covering property restrictions(problematic to express in DL)

Slide 21

Ranking Service DescriptionsRanking Service Descriptions

• Ranking based on Partial Subsumption

(SpA ⊓ Sr)I

• DL Inference

KB ∪ Dr ∪ DpA ∪ DpB ⊨ (SpA

⊓ Sr) ⊑ (SpB ⊓ Sr)

⇒ DpA ≼ DpB

(SpB ⊓ Sr)I

(SpA)I (SpB

)I

Slide 22

ConclusionConclusion

• Provided an intuitive semantics for formalService Descriptions based on Service Instances

• Emphasized the meaning of variance inService Descriptions

• Mapped intuitive notions to formal elements in DL

• Investigated different DL inferences for matching Service Descriptions

• Showed how variance can be treated during matching

• Proposed a ranking mechanism based on partial subsumption of Service Descriptions