Apparatus and method of semantic service correlation system

Abstract

A correlation-based service mediator consists of Semantic Query Generator, Service Query Manager, Service Invocation Flow Generator, Service Invocation Manager, Expression Interpreter and Service Invocation Result Cache. The Semantic Query Generator converts service request to Semantic Query based on input/output parameters in the service request. The Service Query Manager interfaces with an Ontology Engine, passing query results to Service Invocation Flow Generator, which creates service invocation flow definition based on passed query results. The Service Invocation Manager executes flow definitions. It also manages a service invocation result cache, so that invocation results can be saved and reused for later service request. When the execution results are not available in cache, the Service Invocation Manager invokes service according to flow definition. When attribute dependent functions are invoked Expression Interpreter compute the execution results. The service mediator possesses two major functionalities: service matching and service invocation.

Description

BACKGROUND OF THE INVENTION

Field of the Invention

The present invention generally relates to service mediation for supporting interactions among services in heterogeneous and dynamic environments and, more particularly, to a semantic service mediation system that performs service correlation systematically as part of the service mediation, freeing programmers from understanding extraordinary details of service interfaces when enabling service composition.

Background Description

Service mediation is a very active area of research and development. As background to the invention, we first review some work in the area of service discovery (matching), and then look at some service composition prototypes.

Service discovery and matching is one of the cornerstones for service mediations. Current Web service infrastructures have limitations on providing flexibility to choose selection criteria along multiple dimensions. For instance, UDDI (Universal Description, Discovery and Integration) provides limited search facilities that allows only keyword-based searching of services. To overcome this limitation, semantic technology (as described, for example, in B. Benatallah, M.-S. Hacid, A. Leger, C. Rey, and F. Toumani., “On automating web services discovery”, The VLDB Journal, 14(1):84-96, 2005, and M. Paolucci, T. Kawamura, T. Payne, and K. Sycara. “Importing the Semantic Web in UDDI”, Proceedings of Eservices and the Semantic Web Workshop, 2002) is used to support multiple dimension searching criterions for services. For example, in the paper by M. Paolucci et al., the service description capabilities within DAML-S are mapped into UDDI records, in which semantic descriptions are used to support service discovery and matching. In the paper by B. Benatallah et al. a flexible matchmaking among service descriptions and requests by adopting Description Logics (DLs). However, most of these semantic solutions focus on one-to-one matchings.

Typically, a service mediation system contains three roles: (1) service providers, who publish services; (2) service consumers, who request services, (3) service mediators, who are responsible for service repository management, service matching, service invocation and invocation result delivery. The early service mediations are keyword and value-based: (i) the service discovery is keyword-based (e.g., UDDI (Universal Description, Discovery and Integration)); (ii) service invocations are based on the value of exchanged messages, and the mediator does not perform any data transformations during which. For example, a service request is about retrieving a sports car's insurance quote, where the input parameter's type is SportsCar and output parameter's type is CarPremium. For the value-based service mediation, only the services that exactly match input parameter type SportsCar and output parameter type CarPremium can satisfy the request. In case the service request and service interfaces' input/output parameter types are not exactly matched, then the data format transformation needs to be provided by programmers.

Consequently, as an improvement to keyword and type-based solutions, semantics are introduced into service mediations, wherein ontologies enable richer semantics of service descriptions and more flexible matchings. See, for example, B. Benatallah et al., supra, and M. Paolucci, T. Kawmura, T. Payne, and K. Sycara, “Semantic Matching of Web Services Capabilities”, First International Semantic Web Conference, 2002. However, in current semantic service mediation systems, the concept mapping (i.e., A “is a” B) is provided when the service requests and service interfaces are not exactly matched. However, it does not support the mapping that involves transformation functions (e.g., A=ƒ(B)). Therefore, when composting services (as described, for example, in L. Zeng, B. Benatallah, H. Lei, A. Ngu, D. Flaxer, and H. Chang, “Flexible Composition of Enterprise Web Services”, Electronic Markets—The International Journal of Electronic Commerce and Business Media, 2003, and L. Zeng. B. Benatallah, A. H. H. Ngu, M. Dumas, J. Kalagnanam, and H. Chang, “QoS-Aware Middleware for Web Services Composition”, IEEE Transactions on Software Engineering, 30(5):311-327, 2004), developers need to not only understand detail specifications of available service interfaces to create composition schemas, but also implement the data transformation functions.

SUMMARY OF THE INVENTION

According to the present invention, we present the design and implementation of a novel semantic service mediation system which supports not only one-to-one service matchings, but also multiple-to-one service correlations. With our correlation-based mediation, when either adopting knowledge-based or process-based service composition mechanisms, developers can now focus on high level business logic to develop composition services, without understanding extraordinary details of service interfaces. Further, the attribute dependence based correlations perform data transformations systematically, which frees developers from the implementations of data transformation functions.

In our service mediation, semantic information in service descriptions and requests enables one-to-multiple service matchings, which initiates a type of automatic service correlation. Our service correlation is different from the existing industrial and academic service composition frameworks (e.g., J. Koehler and B. Srivastava, “Web service composition: Current solutions and open problems”, ICAPS '03 Workshop on Planning for Web Services, June 2003). The industrial solution typically does not provide explicit goals of the composition and does not describe the pre- and post-conditions of individual services. A service is viewed as a remote procedure call. A service composition is quite often specified as a process model (e.g., BPEL4WS (Business Process Execution Language for Web Services)) though a richer process specification is needed. Ultimately, a process language specification of a Web service composition should contain control-flow descriptions including branching and iteration/recursion for alternative composition execution, dataflow descriptions of the type hierarchy of process artifacts, exception handlers for increased reliability and fault-tolerance of the composition. The messages between the services are simple syntactic descriptions without any semantics specifications. The composition itself is mostly done manually by IT specialists in an ad-hoc manner. In our approach, a collection of services are correlated based on the semantics of service interfaces, without much programming efforts involved.

The semantic web approach, mostly used in the research community relies on the specification of semantics of operations, explicit specifications of goals of composition, pre- and post-conditions of the composed services in a common service ontology, and specification of conditions in temporal logic. A planning algorithm is often used to produce a composite service (see, for example, S. Narayanan and S. McIIraith, “Simulation, Verification and Automated Composition of Web Services”, Proceedings of the 11th International World Wide Web Conference, Honolulu, USA, 2002, and M. Pistore, P. Traverso, and P. Bertoli, “Automated composition of web services by planning in asynchronous domains”, ICAPS '05, 2005). Typically, an AI planning produces a composite service consisting of atomic actions without a hierarchy and contextual information. In our approach, instead of defining the service composition knowledge, the common ontology used in our solution is a general ontology, i.e., it focuses concepts and dependence among them only. Based on the input/output parameters of service interfaces, we compose microflows to correlation services, which is transparent to service compositions.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, aspects and advantages will be better understood from the following detailed description of a preferred embodiment of the invention with reference to the drawings, in which:

FIG. 1 is a diagram of a dependence tree of the class C;

FIG. 2 is a block diagram illustrating the major components of the semantic service mediation system according to the invention;

FIG. 3 is a block diagram illustrating the OWL ontology definition for dependence function;

FIG. 4 is a simplified ER block diagram of the service repository;

FIG. 5 is the service mediator system architecture block diagram according to the present invention;

FIG. 6 is an illustration of the sequence microflow;

FIG. 7 is an illustration of the split-and-correlation microflow;

FIG. 8 is an illustration of the spilt-and-computation microflow;

FIG. 9 is an illustration of the microflow for refined key-based correlation; and

FIG. 10 is an illustration for the microflow of attribute-dependence correlation.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION

In this section, we first introduce some important concepts in ontology, and then present the proposed system architecture of the semantic service mediation. Finally, we present the details of service correlation.

Ontology

In our system, we adopt an object-oriented approach to defining ontologies, in which the type is defined in terms of classes (See Definition 1 below) and an instance of a class is considered as an object (See Definition 2 below). In this subsection, we present a formal description of class and object. It should be noted that this ontology formulation can be easily implemented using OWL (a Web Ontology language). We will present details on how to use ontology to perform semantic matchings and correlation matchings in following sections.

Definition 1 (Class). A class C is defined as the tuple C=<N, S, P, R, F>, where

• • N is the name of the class;
  • S is a set of synonyms for the name of class, S={s₁, s₂, . . . , s_n};
  • P is a set of properties, P={p₁, p₂, . . . , p_n}. For p_iεP, p_i is a 2-tuple in form of <T, N_p>, where T is a basic type such as integer, or a class in an ontology, N_p is the property name. p₁(p₁εP) is the key property for identification;
  • R is a set of parent classes, R={C₁, C₂, . . . , C_k};
  • F is a set of dependence functions for the properties, F={ƒ₁, ƒ₂, . . . , ƒ₁}. Each function is in form of a collection of expressions {ƒ_j1, ƒ_j2, . . . , ƒ_jn}, one for each non-key property, and a predicate c for the key property. And ƒ_jk can be further denoted as p_k=ƒ_jk (p′₁, p′₂, . . . , p′_m), where the p′_i is property from a class other than C and the predicate c is used to correlate p′_i.

In the definition of class, the name, synonyms, and properties present the connotation of a class; while parent classes and dependence functions specify relationships among the classes, i.e., present the denotation of a class. A class may have parent classes for which it inherits attributes. For example, class sportsCar's parent class is Car. Therefore, the class sportsCar inherits all the attributes in class Car.

Other than inheritance relationships, different classes may have value dependence on their properties. In our framework, dependence functions are used to indicate the value dependence among the different classes. For example, we have three classes Duration, Arrival and Departure. In Duration, a dependence function consists of two expressions: {Duration.duration=minus (Arrival.timeStamp, Departure.timeStamp), Duration.unitOfDuration=minute}, where the predicate is Duration.shippingID=Arrival.shippingID=Departure.shippingID.

Based on dependence functions, a dependence tree can be constructed for each class. Assuming that the class C has a set of dependence functions F, a dependence tree can be generated as in FIG. 1. There are three kinds of nodes in a dependence tree, namely class node, operator node and dependant class node. A class C's complete dependence set (denoted as Θ_C) is defined as a collection of depended classes that can be used to calculate the value of the property. For example, the set {C_l, C₁₂, . . . , C_1in} in FIG. 1 is a complete dependence set.

Definition 2 (Object). An object o is a 2-tuple<N_c,V>, o is an instance of a class C, where

• • N_c is the class name of C;
  • V={v₁, v₂, . . . , v₁}, are values according to the attributes of the class C. For v_iεV, v_i is a 2-tuple in form of <N_p, V_p>, where N, is the property name, V, is the property value.Service Interface and Service Request Description

A service interface is denoted as I_s (P_in, P_out), where P_in (P_in=<C₁, C₂, . . . , C_n>) indicates input parameter classes, P_out (P_out=<C₁, C₂, . . . , C_m>) indicates output parameter classes. An example of a service s's interface can be I_s (P_in<SportsCar>, P_out<CarInsurance, CarFinance>), which contains one input parameter and two output parameters.

A service request usually includes functional and non-functional requirements. In this paper, we focus on functional requirements only. A service request is denoted as Q(O_in, E_out), where O_in (O_in=<o₁, o₂, . . . , o_n>) indicates input objects, E_out (E_out=<C₁, C₂, . . . , C_m>) indicates expected output parameters from the services. An example of a service request can be Q (O_in<car>, E_out<CarInsurance, CarFinance>), which contains one input object car and expects a service provides two outputs: CarInsurance and CarFinance.

TABLE 1
Examples
Entity              Example
service request     Q1(Oin   sportsCarA   , Eout   CarInsurance,
                    CarFinance   )
candidate service's I2(Pin   Car   , Pout   CarInsurance, CarFinance   )
interface
interface set       Γk = {I1, I2}, where I1 (Pin   Car   ,
                    Pout   CarInsurance   ), I2 (Pin   Car   ,
                    Pout   CarFinance   )
interface set       Γf = {I1, I2, I3, I4}, where I1(Pin   SportsCar   ,
                    Pout   Licence   ), I2(Pin    Licence   ,
                    Pout   CarOwner   ), I3(Pin   CarOwner   ,
                    Pout   CarInsurance   ), I4(Pin   SportsCar   ,
                    Pout   CarFinance   )
service request     Q2(Oin   shippingPkgA   , Eout   Duration,   )
interface set       Γa = {I1, I2}, where I1(Pin   ShippingPkg   ,
                    Pout   Departure   ), I2 (Pin   ShippingPkg   ,
                    Pout   Arrival   )

Correlation-Based Service Matching

In our framework, a collection of service interfaces be correlated to one that can provide all the necessary outputs required by a service request. Correlation can be either based on common fields and/or attribute dependence functions. For example, two service interfaces I₁ and I₂ in Γ_k (see Table 1 above) can be correlated as they both have the field Car as the input parameter, a key-based correlation service interface set, i.e., The formal definition of a key-based correlation interface set is shown as follows.

In the following each subsection, a various definitions of service interface sets are discussed, wherein the service interface that can be correlated under different conditions.

Parameter-Based Correlation

Obviously, multiple service interfaces can be correlated if they share some input parameters and have different output parameters. Here, we start with the most rigid correlation, where a set of service interfaces that are correlatable by a key input parameter that is specified by the service request.

Definition 3 (Key-based Correlation Interface Set Γ_kc). Γ={I₁, I₂, . . . , I_n} where I_i is a of service interfaces. Γ is a Key-based Correlation Service Interface Set of Q iff:

• • 1. Q can semantically provide all inputs for Γ;
  • 2. Service interfaces in Γ can semantically provide all the outputs expected by Q;
  • 3. Service interfaces in Γ are correlatable for key object o_k;
  • 4. Any service interfaces in Γ contributes a unique output expected by Q.

In this definition, both condition 1 and 2 are necessary conditions, while condition 3 and 4 are the sufficient conditions. Using the above example, the aggregation of I₁ and I₂ provides all the required outputs for the service request, which satisfy condition 1; and their input can be provided by the service request, which satisfies condition 2. Both interfaces have the input parameter Car that is the ancestor of SportsCar—the key class in service request Q. Therefore, the condition 3 is satisfied. Also, I₁ (resp. I₂) provides unique output CarInsurance (resp. CarFinance), which satisfies condition 4. Therefore, I₁ and I₂ compose a key-based correlation service interface set for the service request.

It should be noted that the condition 1 in Definition 3 assumes that there is not any dataflow among the services in the set. We can have a more general definition on the key-based correlation interface set if dataflows are allowed, wherein dataflows indicate that the some interfaces in the set need to be invoked in a sequence. By introducing dataflow, the condition 1 is refined as:

For any input required by the service interfaces in F either

• • is provided by Q or
  • is provided by outputs of a service interface I_i in Γ.If the input is appeared in input parameters of interface I_j, then interface I_j is invoked after I_i and does not need to provide any inputs for interface I_i and any other interfaces that invoked before I_i. Actually, by further releasing the constraint that correlating is based on input key field only, more generic correlations can be defined. For example, the correlation can be based on any input objects. In such a generic correlation, we adopt the notion of Correlatable Class (see Definition 4). In this definition, two classes are correlatable in a set of interfaces if either they appear in the same interface's inputs, or when these two classes do not appear in the same service interface (belong to two interfaces I_x and I_y respectively), either (i) I_x provides at lease one input parameter for I_y (other input parameters can be provided by the service request) or vice versa; or (ii) I_x and I_y share some input parameters, or (iii) there is a sequence of interfaces in the set that are correlatable “step by step” and aiming for correlating I, and I, in the end. Actually, if we consider I_x and I_y are entities in ER model, then these services between I_x and I_y in the sequence are relationships: in order to join two entities without common attributes, a collection of relationships [I_x+1, I_x+2, . . . , I_y−1] are required. For example, class SportsCar and CarInsurance are correlatable in I_f (see Table 2), as class SportsCar and CarInsurance appear in I₁ and I₃ respectively; and 12 is considered as a relationship to bridge SportsCar and CarInsurance.

Definition 4 (Correlatable Class). Class C_i, C_j are correlatable in an interface set Γ (Γ={I₁, I₂, . . . , I_n}), iff either

• • C_i and C_j appear in same service interface; or
  • C_i and C_j do not appear in a service interface, then ∃I_x, I_yεΓ, x≠y, C_i (resp C_j) is output class of I_x (resp. I_y) and either:
    • I_y can provide some input for I_x; or
    • I_x can provide some input for I_y; or
    • I_x and I_y have some comment input; or
    • there is an interface sequence [I_x, I_x+1, I_x+2, . . . I_y−1, I_y] in Γ, for any I_i, I_i+1 in the sequence, I_i's outputs provide some inputs for I_i+1, and for each I_i in the sequence, input parameters can be provided either by the I_x, I_x+1, . . . , I_i−1 or the service request.

Based upon the notion of correlatable class, we can define the concept of Parameter-based Correlation interface Set (see Definition 5 below). Different from key-based correlation, this definition allows correlations on any fields.

Definition 5 (Parameter-based Correlation Interface Set Γ_pc). Γ={I₁, I₂, . . . , I_n}, I_i is a service interface, F is a Parameter-based Correlation Interface Set of service request Q if:

• • 1. Q can semantically provide all inputs for Γ;
  • 2. Service interfaces in F can semantically provide all the outputs expected by Q;
  • 3. Any outputs in service interfaces in F are correlatable;
  • 4. ∀I_iεΓ, at lease one of the following is true:
    • contributes at least one unique output expected by the service request;
    • appears in interface sequences in condition 3 of this definition if some outputs are correlated by interface sequences.Attribute-Dependence Correlation

Other than parameter-based, multiple interfaces can be correlated using dependence functions. Such cases happen when some required output parameters can not be provided by any available interfaces. Assuming that an absent parameter's class C_i has a dependence function, the service mediator can compute the value of the absent output parameter using the attribute-dependence function. For example, if the class type Duration is required by the service request but is not provided by any services, as Duration's dependence set is {Departure, Arrival}, the system can search services that have output Departure or/and Arrival and correlate these output and compute the value for Duration. By generalizing this example, we can propose the definition of Key-based Attribute-dependence Correlation Interface Set (See Definition 6 below). Again, we first limited the correlation on key field only, wherein can be defined as:

Definition 6 (Key-based Attribute-dependence Correlation Interface Set Γ_ka). Γ={I₁, I₂, . . . , I_n}, I_i is a service interface. Γ is a Key-based Attribute-dependence Correlation Service Set of the service request Q iff:

• • 1. Q can semantically provide all inputs for Γ;
  • 2. For any outputs expected by the service request, either
    • a service interface I_i can provide a semantic compatible output; or
    • outputs of service interfaces contain a complete dependence set Θ_Di for the output required by the service request.
  • 3. Service interfaces in Γ are correlatable for key object o_k;
  • 4. Any service interfaces in Γ contributes a unique output expected by Q or Θ.

In condition 2 of the above definition, unlike the definition of parameter-based correlation interface set, a parameter required by the service request may not appear in any services. However, it can be computed using dependence functions (See Definition 2 above). Like parameter-based correlation interface set, the condition 3 concerns whether interfaces can be correlated by the key field. An example of key-based attribute-dependence correlation service set is Γ_a, for the service request Q₂. In the example, the request output Duration is not directly provided by any interfaces. Instead, two interfaces I₁ and I₂ provide outputs Departure and Arrival respectively, and Duration is then computed based on them.

Again, we can release the constraint that correlations are based on a key-field only. Therefore, the more generic Attribute-dependence Correlation Interface Set can be defined (see Definition 7 below). In particular, the condition 3 of the definition indicates that correlation can be done based on any fields.

Definition 7 (Attribute-dependence Correlation Interface Set Γ_ac). Γ={I₁, I₂, . . . , I_n}, I_i is a service interface. Γ is a Attribute-dependence Correlation Interface Set of service request Q iff:

• • 1. Q can semantically provide all inputs for Γ;
  • 2. For any outputs expected by the service request, either
    • a service interface I_i can provide a semantic compatible output; or
    • outputs of service interfaces contain a complete dependence set Θ_Ci for the output required by the service request.
  • 3. Any outputs in service interfaces in Γ are correlatable;
  • 4. Any service interfaces in Γ contributes a unique output expected by Q or Θ.

As shown in FIG. 2, our semantic service mediation consists of three main components: ontology repositories 10, an ontology engine 11 and a service mediator 13. The ontology engine 11 accesses the ontology repositories 10 and communicates with the service mediator 13. The service mediator 13 accesses the service repository 12 and responds to service requests from service requestor 15 and invokes service providers 14 who publish services, as will be described in more detail below.

Ontology Representation

Most of the object-oriented ontology notations (see Definition 1 and 2) can map to OWL DL constructs (See Table 2), except for dependence function. Basically, dependence functions describe relationship among classes. Expressing such relationship requires OWL FULL instead of OWL DL, if we use only one OWL ontology to represent one object-oriented ontology. However, OWL FULL ontology is proven to be undecidable. In order to comply with OWL DL, two OWL ontologies are used. O_n is used to present the ontology without dependence functions, while O_d (See FIG. 1) represents dependence function among classes in O_n. Representation of dependence function in OWL is given in Table 3. In our setting, the annotation properties, rdf:seeAlso, is used to link O_d and O_n.

TABLE 2
Map Object-Oriented Notation to OWL
Object-
Oriented
Ontology     OWL DL Construct
Class C      owl:Class
Class        rdf:about
name N
Synonyms S   owl:equivalentClass, e.g. <C, owl:equivalentClass, S1>
Properties P owl:DatatypeProperty (rdfs:range is primitive type), or
             owl:ObjectProperty (rdfs:range is a class in OWL ontology)
Subclass R   rdfs:subClassOf, e.g.<C, rdfs:subClassOf, C1>
Object o     owl:Individual
Value V      RDF triple, e.g. <i, p, v>.
             the value of property p of individual i is v.

For example, Duration.duration=minus (Arrival.timeStamp, Departure.timeStamp) is a dependence function, and Duration is the DefinedClass. This dependence function has two DependenceClass, Arrival and Departure. It has one DependenceExpression, where operator is minus and DefinedProperty is Duration.duration. The DependenceExpression has two DependenceProperty, Arrival.timeStamp and Departure.timeStamp. The mapping between dependence function in OWL and its Object-Oriented representation could be found at Table 3.

It should be noted that OWL ontologies in this invention are developed and visualized by using EODM RSA Workbench in IBM Integrated Ontology Development Toolkit (IODT).

TABLE 3
Representation of Dependence Function in OWL
Object-Oriented Ontology Dependence Function in OWL
Dependence function fi   DependenceFunction
Expression fik           DependenceExpression
Expression output pk     DefinedProperty
Expression operand pi′   Dependence Property
Target class             DefinedClass
Source class             DependenceClass
Predicate c              DependenceFunction.predicate

TABLE 4
Query Services
Basic Query Services             SPARQL Query
getSubClass (C):                 PREFIX sme: <http://service.ibm/example#>
get all children classes of class C SELECT ?c
                                 WHERE {?c rdfs:subClassOf sme:C}
getEqClass (C):                  PREFIX sme: <http://service.ibm/example#>
get all synonym of class C,      SELECT ?c
including reasoning of owl:equivalentClass WHERE {?c owl:equivalentClass sme:C}
getDirectDependeceSet(C):        PREFIX sme: <http://service.ibm/example#>
get direct dependence set of class C PREFIX ds: <http://service.ibm/dependence#>
                                 SELECT ?c
                                 WHERE {sme:C rdf:seeAlso ?x . ?x ds:hasFunction ?f . ?f ds:hasClass ?c}
getDependenceFunction(C):        PREFIX sme: <http://service.ibm/example#>
get dependence function of class C PREFIX ds: <http://service.ibm/dependence#>
                                 SELECT ?f
                                 WHERE {sme:C rdf:seeAlso ?x . ?x ds:hasFunction ?f}

Ontology Query Service

With the above OWL presentations, we can use OWL ontology repository to provide ontology query services, wherein two OWL constructs rdfs:subClassOf and owl:equivalentClass in O_n trigger OWL reasoning and produce inferred facts. In our implementation, most of the OWL ontology repositories are doing reasoning at loading time and all the inferred results are stored in the repository as well. After loading two ontologies, O_n and O_d, a set of ontology query services are defined to retrieve combined ontology information by issuing SPARQL queries (Query Language for RDF, a directed, labeled graph data format for representing information in the Web). Basically, the ontology query services can be categorized into two categories, namely basic query service and dependence query service. The basic query services are providing methods to retrieve basic ontology information like class, property and individual. They can be achieved by querying O_n only. The dependence query services must be implemented by issuing combined query over O_n and O_d, to provide access to dependence set and dependence functions of a specific class. Some basic query services and associated SPARQL queries are given in table 4. In particular, getDirectDependenceSet (C) can only retrieve direct dependence set of class C. An algorithm to retrieve the complete collection of dependence sets of class C could be easily implemented by using the Breadth-First Traversal algorithm to traverse the instance graph of O_d.

Service Repository

We adopt a type-instance approach to organize the service repository: the service interface's input/output parameters are used to identify types of interfaces, while the information (in service publication) such as invocation, QoS, etc., are used to identify instances of service interfaces. In our solution, the service repository separates the data type information and contents of service interface (see FIG. 3 for simplified data schema), wherein one table Classes (T for short) is used to store the class type information for parameters in service interface, and another table Parameters (P for short) is used to store the parameter information of interface types. It should be noted that both class information and the content of the input/outputs are stored vertically in these tables. In particular, for table Classes, each class used in a service interface occupies a row. For each type of interface, a unique interfaceTypeID is assigned in table interfaceTypes (I for short). For each interface instance, a unique interfaceInstanceID is assigned in table interfaceInstance (S for short) and associated with interface type using the field interfaceTypeID. It should be noted that for the sake of presentation, we do not present the details of service description other than the service interface signature.

Service Mediator

The service mediator 13 is shown in FIG. 5 and consists of Semantic Query Generator 51, Service Query Manager 52, Service Invocation Flow Generator 53, Service Invocation Manager 54, Expression Interpreter 55 and Service Invocation Result Cache 56.

Semantic Query Generator 51 handles service requests. It generates Semantic (e.g., SPARQL) Query statements, wherein the detail algorithm can be found in the Ontology Query Service Section. Basically, the generator converts service request to Semantic Query based on input/output parameters in the service request.

Service Query Manager 52 interfaces with Ontology Engine 11, in order to execute the Semantic Queries. It will pass the query results to Service Invocation Flow Generator 53.

Service Invocation Flow Generator 53 creates service invocation flow definition based on query results passed from Service Query Manager 52. The details about flow generation are discussed below in Section on Service Invocation.

Service Invocation Manager 54 executes the flow definitions. It also manages a service invocation result cache 56, so that invocation results can be saved and reused for later service requests. When the execution results are not available in cache, the Service Invocation Manager 54 invokes service according to flow definition. It should be noted that when attribute dependent functions are invoked, the Expression Interpreter 55 computes the execution results. The service mediator 13 possesses two major functionalities: service matching and service invocation.

Service Matching

We discuss generation of queries for searching for a single service that can match a service request first. Basically, there are two steps involved: (i) generating queries to search all the service interface types that contain all the semantic compatibility output parameters for the service request; (ii) generating queries to inspect whether the service interface can be invoked using the inputs provided by the service request. In the first step, assuming the service request is Q(O_in, E_out), m (m is the number of expected outputs) queries are generated to search interface types that provide semantically compatible output parameters in E_out:ρ(ƒ_i,σ_{T,classNameεgetSubClass(Ci.className)ΛT,isInput=false}T_classIDP),iε[1 . . . m]  (1)

Because the parameter information is stored vertically, equijoining ƒ₁ on interfaceTypeID is required, in order to verify whether the interface type can provide all the compatible parameters required by the service request.ρ(ƒ_out,_{interfaceTypeID}{ƒ_i,iε[1 . . . m]})  (2)

Now, ƒ_out presents the all the interface types that can provide all the necessary output parameters for the service request. The next step is to determine whether the services can be invoked by the inputs provided by the service request Q in O_in.ρ(ƒ,σ_{(minus(πinterfaceTypeIDƒout,COin)=ø)}ƒ_out)  (3)

In the query, Ω_Oin is set of classes that contains all the input objects in the service request, and the function minus (Set Ω₁, Set Ω₂) finds out all the classes in Ω₁ that can not find semantic compatible class in Ω₂. Therefore, if minus (π_{ƒout.interfaceTypeID}, Ω_Oin) returns null, it indicates that all the required inputs can be provided by the service request.

Now we discuss how to generate queries to search key-based correlation interface sets. Assuming that in the service request Q, the key's class type is C_k, two queries are generated for each output parameter class C_i:ρ(ƒ_k,(σ_{Ck,className=classNameT,inInput=true}T_classIDP))  (4)ρ(ƒ_i,(σ_{Ci,classNameεgetSubClass(className)T,isinput=false}T_classIDP))  (5)where query ƒ_k searches service interface types that use C_k as an input parameter and query ƒ_i searches service interface that contribute an output parameter C_i. For example, in Q, the key field's class is SportsCar. For the output parameter CarInsurance, two queries are generated as:ρ(ƒ_k,(σ_{T,className=′SportsCar′T,isInput=true}T_parameterIDP))  (6)ρ(ƒ_i,σ_{Ci,className=′CarInsurance′)T,isInput=false}T_parameterIDP))  (7)By equijoining ƒ_k and ƒ_i on interfaceTypeID, we have ƒ_k,i (see query 8) that represents interface types that contain both key parameter C_k as input and C_i as output, in which the interface types satisfy the condition (2) and (3) in Definition 3.P(ƒ_k,i,ƒ_{kinterfaceTypeID}ƒ_i)  (8)

Using the above example, ρ(ƒ_k,1, ƒ_{k interfaceTypeID}ƒ₁) is generated for searching service interface types that contain both SportsCar as input parameter and the output parameter that is semantically compatible with field CarInsurance. Now, we discuss generating queries to search interface type that can satisfy the condition (1) in Definition 3.ρ(ƒ′_k,i,σ_{(minus(πinterfaceTypeIDƒk,i},Ω_Oin)=Ø)ƒ_k,i)  (9)

By joining all the ƒ′_k,i, we have the query ƒ′_k,1ƒ′_k,2 . . . ƒ′_k,m that gives all possible key-based correlation interface set. In case any ƒ′_k,i returns null, which indicates the expected output parameter C_i is missed. In such case, the service mediator can search attribute-dependence correlation set. Assumes that Θ_Ci (Θ_Ci={C₁, C₂, . . . C_l}) is complete dependence set for class C_i, If we consider each class in Θ_Ci expected output parameter class, same queries as (8) can be generated as an output parameter.

Service Invocation

Once a correlation interface set is identified, the service mediator constructs a microflow (represented as statechart) to compose services as a “virtual service” for the service requestor. Based on the input/output dependence, correlation relationship and attribute-dependence functions, there are three basic patterns in generating microflows.

• • 1. Sequence. In this case (see FIG. 6), interfaces in a sequence [I_x, I_x+1, . . . , I_y] have input/output dependence. In particular, for any I_i in the sequence, it expects some outputs from some interfaces in [I_x, I_x+1, . . . , I_i−1], while it does not provide any input for the interface in [I_i+1, . . . , I_y].
  • 2. Split-and-Correlation. In this case (see FIG. 7), interfaces in a set {I_x, I_x+1, . . . , I_y} do not have any input/output dependence and the execution results need to be correlated and transformed to the format required by the service request. Therefore, in the microflow, these interfaces are invoked in paralleled and then a correlation task t, is performed.
  • 3. Split-and-Computation. In this case (see FIG. 8), like previous case, interfaces in a set {I_x, I_x+1, . . . I_y} do not have any input/output dependence. Different from the previous case, these interfaces' outputs consist of a complete dependence set for a missing output parameter that is required by the service request. Therefore, after these interfaces are invoked in parallel, a computation task t_ƒ is then performed by the service mediator to compute the value of the missing output.It should be noted that any I_i in the above three patterns may not be an interface provided by the service provider, it can be provided from a “virtual service” composed by a microflow pattern.

Now we discuss how to use the above three patterns to generate microflows for correlation interface sets. For the key-based correlation interface set (see Definition 3), the microflow is generated using split-and-correlation (see FIG. 7), as all the interfaces in the set do not have any input/ouput dependence. For the refined key-based correlation interface set that allows input/output dataflow among the interfaces, the microflow is generated using two constructs: sequence and split-and-correlation. An example of microflow for refined key-based correlation service set is shown in FIG. 9. It should be noted that, some interfaces in the set have input/output dependence. Assuming these interfaces compose a sequence [I_x, I_x+1, . . . I_y] (multiple sequences may created), they need to be invoked in order.

For the attribute-dependence correlation interface set (see Definition 7), all three patterns are applied to generate related microflows. An example microflow for attribute-dependence correlation is shown in FIG. 10. In the example, interfaces in the sequence [I_x, I_x+1, . . . , I_y] that are the elements in the set that are used to correlate parameters that are not appeared in the same interface. While interfaces in set {I′_x, I′_x+1, . . . , I′_y} provide complete dependence set for a missing output they do not have any input/output dependence.

In case multiple interface sets are identified for a service request, then multiple microflows are generated. In such cases, a quality-driven approach is adopted to select the best quality microflow for the service request. Once the microflow is generated, the service mediator orchestrates the execution of the services based on the control-flow and dataflow. Further, the service mediator possesses a computation engine for executing the attribute-dependence functions defined in the ontology. We implemented the computation engine on top of a XPath 2.0 expression engine.

While the invention has been described in terms of a single preferred embodiment, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims.

Claims

1. A computer implemented method of service correlation, comprising the steps, performed by a computer in a semantic service mediation system, of: receiving a service request that provides one or more input parameters and indicates expected output parameters, wherein the one or more input parameters and the expected output parameters include objects according to an object-oriented ontology, wherein an object is defined as an instance of a class;searching a service repository of available services to identify a candidate service interface set as being a key-based attribute-dependence correlation service interface set that satisfies the service request received in the receiving step, wherein the candidate service interface set is a plurality of service interfaces corresponding to some of the available services in the service repository, wherein the searching step is performed based only on input and output parameters of service interfaces without describing pre- and post-conditions of individual services, and wherein the searching step includes, for the candidate service interface set, querying whether for all input parameters for the candidate service interface set, each input parameter is either included in the one or more input parameters provided by the service request or else is provided as an output of a service interface in the candidate service interface set,querying whether all of the output parameters expected by the service request are provided by the candidate service interface set, including querying whether one or more output parameters expected by the service request are not provided by any individual service interface of the candidate service interface set, andquerying whether outputs of the plurality of service interfaces contain a complete dependence set for each of the one or more expected output parameters that are not provided by any individual service interface, wherein the complete dependence set is a collection of depended classes used together to arithmetically calculate a value for an output parameter,querying whether the plurality of service interfaces forming the candidate service interface set each require an input parameter that is an object of a key class, including selecting, from classes corresponding to the one or more input parameters of the service request, one class for designation as the key class, andfor each service interface in the candidate service interface set, comparing each input parameter of each service interface for a match to the designated key class at least until a match is found or all input parameters of each service interface have been compared,querying whether each of the plurality of service interfaces forming the candidate service interface set contributes a unique output parameter either for the expected output parameters expected by the service request or for the complete dependence set, andidentifying the candidate service interface set as being the key-based attribute dependence correlation service interface set if and only if all of the querying steps return true,wherein a service interface indicates one or more input parameters required by a corresponding service and indicates one or more output parameters provided by the corresponding service, andwherein each service interface of the plurality of service interfaces is alone insufficient to satisfy the service request received in the receiving step;generating a microflow from the plurality of service interfaces forming the key-based attribute-dependence correlation service interface set; andexecuting the microflow generated in the generating step to invoke services from the available services in the service repository to satisfy the service request.

2. The method of claim 1, further comprising steps of translating the service request to a semantic query, and using semantic information in service descriptions of the available services in the service repository and in the semantic query, performing one-to-multiple service matchings.

3. The method of claim 2, wherein the searching step further includes executing the semantic query to locate candidate service providers for the service request.

4. The method of claim 1, wherein the searching step further comprises identifying one or more further candidate service interface sets as correlation interface sets based on one or more of parameter-based correlation and attribute-dependence based correlation; wherein service interfaces in parameter-based correlation interface sets are correlatable for any common input parameters, andservice interfaces in attribute-dependence correlation interface sets are correlatable for any common input parameters.

5. The method of claim 4, wherein the searching step, using the parameter-based correlation, further comprises the step of locating candidate service interface sets that satisfy the following criteria: inputs for all service interfaces in a given candidate service interface set are provided by the one or more input parameters from the service request;the service interfaces in the given candidate service interface set provide all the output parameters expected by the service request;any outputs of the service interfaces in the given candidate service interface set are correlatable; andany individual service interface in a the given candidate service interface set contributes a unique output expected by the service request.

6. The method of claim 4, wherein the searching step, using the attribute-dependence correlation, further comprises the step of locating candidate service interface sets that satisfy the following criteria: inputs for all service interfaces in a given candidate service interface set are provided by the one or more input parameters from the service request;for any of the output parameters expected by the service request, either a service interface in the given candidate service interface set provides a semantic compatible output, or outputs of service interfaces in the given candidate service interface set contain a complete dependence set for the output required by the service request;any outputs in the service interfaces of the given candidate service interface set are correlatable; andany individual service interface in the given candidate service interface set contributes a unique output expected by the service request or the complete dependence set.

7. The method of claim 1, further comprising a step of supporting one-to-one service matchings.

8. The method of claim 1, wherein the searching step includes identifying additional candidate service interface sets that satisfy the service request, wherein the generating step includes generating multiple microflows for the additional candidate service interface sets, followed by selecting a best quality microflow for satisfying the service request using a quality-driven approach.

9. The method of claim 1, comprising a further step of searching the service repository of available services to identify a second candidate service interface set as being a parameter-based correlation service interface set that satisfies the service request received in the receiving step, wherein the second candidate service interface set is a plurality of service interfaces corresponding to some of the available services in the service repository, and wherein the further searching step includes, for the second candidate service interface set, i. querying whether for all input parameters for the second candidate service interface set, each input parameter is either included in the one or more input parameters provided by the service request or else is provided as an output of a service interface in the second candidate service interface set,ii. querying whether all of the output parameters expected by the service request are provided by the second candidate service interface set,iii. querying whether any outputs in the services interfaces of the second candidate service interface set are correlatable,iv. querying whether each of the plurality of service interfaces forming the second candidate service interface set either contributes at least one unique output parameter for the expected output parameters expected by the service request, orappears in interface sequences in query iii. if some outputs are correlated by interface sequences, andidentifying the second candidate service interface set as being the parameter-based correlation service interface set if and only if all of the four querying steps i. through iv. return true.