290 likes | 298 Views
This paper explores semantic analysis and matching approaches in e-Business service discovery, focusing on intuitive results, OWL-DL, and variance in service descriptions. It discusses service requester mechanisms and automation frameworks that streamline partner discovery and contract negotiations.
E N D
WSML Presentation Variance in e-Business Service Discovery Uwe Keller based on a paper by S. Grimm, B. Motik and C. Preist (to be published) and slides by S. Grimm for the SWWS Evaluation Meeting, Lausanne, France, 2004.
Motivation … Automation of B2B partner discovery and contract negotiation (Project SWWS) First step: • Identify potential business partners that provide suitable services by service / request matching • Many approaches are based on DLs • But sometimes do not produce intuitive results This paper … • Analyzes semantic of service descriptions on intuitive notions • Introduce different kinds of variance in service descriptions • Shows how to exploit that in matchmaking phase • Maps intuitive notions to constructs in OWL-DL • Investigates different inferences and defines semantics of discovery
Service Discovery … A service requester may use a discovery mechanism to locate a set of providers which are potentially able to meet its needs. A framework for automation service discovery requires: • A language for describing services • with formal semantics • which matches the intuitive understanding of modellers • Matching algorithms for implementing discovery In this paper … • Following common approaches: Use Description Logics (OWL-DL) • Despite common approaches: Stick to the intuitive semantics a service modeller expects & provide methodological guidelines • Focus on the problem of variance (in service descriptions)
Services & Service Descriptions • Widespread meanings of the term „service“ • as: Abstract business interaction between two parties • as: Computational entity with a WS interface • Proposed model • Set-based model • Distinguish service instance and abstract service class • Real-world service = instance (Agreed Service) • Represents agreement in all details of a service to be provided between requestor and provider (contract) • Service description = set of (agreed) services = service class • Classes capture variance in provided (and requested) services
Service Descriptions (II) • Proposed model (cont‘d) • In a B2B scenario, this is natural way for providors/requestors to express their capablities/needs: • Both describe the space of possible concrete service instances / contracts which are acceptable for them • Service descriptions • act as templates for contracts • induce variance • Service instances can be represented as … • directed labeled graph / instance of relational schema
is-a Example … • Service instance is-a Shipping Crate
Possible representation:DL concept expression: Capabilitys≡ Shipping ⊓ ∃from.UKCity ⊓ ∃to.GermanCity ⊓item.Container . . . Models variance in acceptable contracts Example (II) … • Service description Service instance / contract 1 Service instance / contract 2
Service Descriptions (III) • Provider and Requestor descriptions are treated in the same way! • Variance in service models: 2 flavours can occur • Variance due to intended diversity in contracts (Service Descr. as concepts) • Variance due to incomplete knowledge (Open-world semantics of DLs) • … and can/should be distinguished (in matchmaking!) • How to reflect variance due to incomplete knowledge? • Consider many possible worlds (each one detailing out the „missing“ information) • Withing each possible world: intentended diversity by set of acceptable contracts
Service Descriptions (IV) • Different kinds of variance … • Incomplete information is (fully) detailed out by selecting one possible world • Intended diversity is represented as many alternative contracts within this world
Matching for Discovery … • Discovery = matching goals and capabilities • checking if goal and capability allow common service instances • If there are common instances, requester and provider can (potentially) do business with each other (Goal)I∩ (Capability)I ≠ Ø
II. Operationalize Discovery via Logics: From Intuition to Logics
From Intuition to Logics … • How to represent the informal elements described before in DL? • Service Description = Set of DL axioms D (using some concept S somewhere) • Domain specific background knowledge = DL knowledge-base KB (containing all relevant facts) • Possible world = Model I of KBD • Contract = Relation structure (Instance with complex properties) • Acceptable contracts = Instances in the extension I(S) of S • Variance due to intended diversity = I(S) is not a singleton set • Variance due to incomplete knowledge = KBD has several models I1, I2, … • Matching = Applying DL-inferences in a procedure match(KB, S-req, S-prov)
From Intuition to Logics … • How to represent typical informal characteristics of service properties in DL? • Variety • Property is fixed to a specific value i or can range over a certain class C: r.{i} or r.C • Availability • Property can be obligatory or optional: r.T • Multiplicity • Property can be multi-valued or single-valued : ≤1r • Coverage • Property can be range covering: In every possible world, for any value in the range C, there is an acceptable contract with this property value: C r-.S
Matching Service Descriptions • Treating Variance due to Incomplete Knowledge • Reflected by multiple models I of KB* • Two ways to reason with • Is there a way of resolving unspecified issues (i.e. possible world) such that the two service descriptions accept a common contract • Satisfaction check KB*{Match(Sr,Sp)} • Regardless of how to resolve unspecified issues, requestor and provider have common contracts in every possible world • Check entailment of Match(Sr,Sp) by KB*
Matching Service Descriptions • Treating Variance due to Intended Diversity • Reflected by multiple alternate contracts within a single model I of KB* • Three ways to reason with • Is there a common contract for both parties • Match(Sr,Sp) := Sr ⊓Sp ⋢⊥ • Requested service is more specific • Match(Sr,Sp) := Sr⊑Sp • Requested service is more general • Match(Sr,Sp) := Sp⊑Sr
Inferences for Matching • (1) Satisfiability of concept conjunction • Weakest check that can be applied (along the 2 dimensions) • One possible world • One common instance
Inferences for Matching • Example • Match(KB, Dr, Dpa) : yes! • Match(KB, Dr, Dpb) : yes!(since UKCity⊓USCity ⋢⊥ is not specified in KB strange!)
Inferences for Matching • (2) Entailment of concept subsumption • Very strong check (in comparison to (1) ) • Regardless of possible worlds (in any of them) • All instances of one service description is a subset of the other service description (provider/requestor)
Inferences for Matching • Example • Match(KB, Dr, Dpa) : no! (Dublin not in the UK) • Match(KB, Dr, Dpb) : no! (Plymouth not in US) • To strong for the general case (Dpa should match)
Inferences for Matching • (3) Entailment of concept non-disjointness • Overcomes deficiencies of (1) and (2) • Regardless of possible worlds (in any of them) • Checks for a common contract in each possible world • Stronger than (1), weaker than (2)
Inferences for Matching • Example • Match(KB, Dr, Dpa) : yes! • Match(KB, Dr, Dpb) : no! (Plymouth not in US) • Intuitive expected result is achieved!
Open issues … • „Standard“-Notion which captures intuition • Entailment of Concept Non-Disjointness • But: Requires modellers to seperate out possible worlds in which some of their intended accepted contracts are missing (All of them have to be captured) • Can be achieved by Range-Covering property restrictions • However: DL has only restricted expressiveness when several properties are involved at the same time • Requires coverage of all possible combinations of values Not expressible in DL But perhaps with DL+Rules ?
Conclusions … • Paper descriped a Service Discovery Framework for the e-Business setting based on DL • Provided semantics to formal service descriptions in an intuitive way • Indentified two dimensions of variance in formal service descriptions • Mapped intuitive notions to formal representatives in DL • Introduced a set of attributes by which properties can be restricted (and how this is done in DL) -> Methodology! • Discussed several inference that can be used for matchmaking along the two dimensions of variation • Proposes Entailment of Concept Non-Disjointness to overcome some deficiencies of the notions that have been proposed in the DL-literature so far. • Proposed new ranking scheme (omitted in this presentation)
Relevance to WSMO … • Model close to parts of D5.1 (WSMO Discovery) • Distinguish service and web service • Discovery != finding a real-world service • Descriptions on the level of simple semantics annotations (Action-Object-Model) • Does not cover the mediation problem • Investigates a specific model for simple semantic annotations by means of an example • Action-Object model • Gives a set of attributes for restricting properties (of a contract) for simpler modelling (Availability, Muliplicity, …) • Variance due to incomplete information is treated in D5.1 only in one way (independence of incomplete information, not sure whether the other alternative is actually useful)
Relevance to WSMO (II) … • Fixes a standard notion (for B2B scenario) • In D5.1 we don‘t do that by purpose • Range coverage and related problems • Is needed for us as well with the intersection match • (= materializing the universe at a logical level) • We can deal with that if we use a expressive language like WSML-FOL (thought this might be more complex if the service (action-object-model) would be more complex).