Automatic Synthesis of Reference Processes Applying Semantic Concepts

Transcription

1 Automatic Synthesis of Reference Processes Applying Semantic Concepts Florian Lautenbacher, Bernhard Bauer Programming of Distributed Systems Institute of Computer Science, University of Augsburg, Germany [lautenbacher Abstract. Optimizing business processes is a necessary prerequisite to reduce transactional costs in and between enterprises. Therefore companies spend huge efforts for adapting predefined processes, modeling their value-added business processes and implementing them with IT systems. Several tools are available for modeling the processes and the generation of code, nevertheless changing a sub-process often requires to be propagated to the whole process. Thus, methods are necessary to support automated actualization of process models omitting this time-consuming manual task. To reach this goal, the synthesis of predefined processes has to be automated. Therefore process models need to be extended with information describing the semantics of the processes. To automate this task Semantic Web Standards can be applied to obtain machineunderstandable information. Extending each process description with semantic information enables an automatic synthesis of value-added processes. This paper shows a first approach how to annotate process models with semantic data, describes synthesis algorithms and evaluates a prototypical implementation. 1 Introduction Over the past few years, enterprises have been undergoing a thorough transformation in reaction to challenges such as globalization, unstable demand, and mass customization. A key to maintain competitiveness is the ability of an enterprise to describe, standardize, and adapt the way it reacts to certain types of business events, and how it interacts with suppliers, partners, competitors, and customers. In the context of process orientation, enterprises today describe these procedures and interactions in terms of business processes and invest huge efforts to describe and standardize them. The processes are either notated only on a textual basis or graphically with models. During the last decades several graphical model standards emerged like the SCOR-model [12], ARIS [13] or the Unified Modeling Language (UML2) [14] which is more established in the section of computer science. In particular, the UML 2.0 standard with its extended activity diagrams supports an elegant modeling of business processes. Business process models can be created and modified with various tools supporting one of these standards. Some companies prefer an UML-designer like MID s innovatoraox [3], others the ARIS toolset [12]. Some of these tools provide the possibility of model transformation or code generation additional to the simple crea-

2 tion and modification of models. But, currently none of these tools are providing an assistance to orchestrate the modeled processes. Currently companies standardize their business processes and create pre-defined business processes, so-called reference processes, which can be adapted to the changing demands of customers, i.e. reference processes can be instantiated or customized in different organization units of an enterprise. Meanwhile, more than a hundred different reference processes are available within an enterprise. These processes can be adapted to every organization unit and their business cases. But an integration of all needed reference processes to the value-added processes of a company and the modification is a time-consuming task. An automatic adaptation of reference processes to the needs of an enterprise or organization unit could increase their utilisability and usage. This requires a description of the processes that machines can interpret. Standards of the semantic web are designed to be machine-understandable and provide the required functionality. This paper describes an approach to annotate business and reference processes with machine-understandable information based on standards of the semantic web and semantic web services. Using these metadata a computer is able to compute a (semi-)automatic orchestration of all processes (called synthesis). Therefore, synthesis algorithms were developed and are surveyed to calculate an optimal composition of all processes (concerning the given parameters). This enables an easy customization of a reference process to the needs of a company or organization unit, because the computer takes care whether the orchestration of all sub-processes has to be adapted, too, or not. This is to our knowledge the first approach to annotate business and reference processes and use these annotations to make an automatic synthesis of pre-defined processes to value-added processes. Other institutions are describing business processes as well: [28] describes an approach to use ontologies for business process modeling in Petri nets. Jenz & Partner is developing a tool named BPEdit [10] which enables users to graphically define and edit business process definitions using BPMN [15]. BPEdit relies on an open and extensible ontology-based information model which is defined in OWL and offers the generation of deployable process definitions in languages such as BPEL. The METEOR-S development team is developing a Process Designer [11] to generate BPEL Web Processes based on the METEOR-S Web Service Annotation Framework [17] which supports dynamic discovery of partner services using semantics. Both tools enable the annotation of business processes and generate process definitions in BPEL. They use semantic web languages and develop concepts to annotate business processes with semantic information. But at present they are not using these annotations to make an automatic synthesis of the described processes. There are several groups working on composition of web services: some of them require static information (e.g. Golog-based [23] or based on Hierarchical Task Networks [24]). Some approaches use ontologies [25], others agent technologies [26]. But none of them currently considers business processes and the automatic composition of process models in UML2 activity diagrams. This paper is organized as follows: Section 2 provides the background. The concepts of these standards are the basis to annotate business process models as de-

3 scribed in section 3. In the same section algorithms for the synthesis of business processes are presented and explained. In section 4 a case study based on a prototype implementation is shown. Finally section 5 summarizes the opportunities of our approach and outlines further research activities. 2 Background In order to obtain a (semi-)automatic synthesis of business processes, processes have to be annotated with semantic information. This requires a language that computers can decode and analyze, i.e. it has to be machine-understandable data. Here the advantages of semantic web standards come into play. Semantic web constructs can be used to describe the data structure as well as the dynamic behavior of an enterprise. This information is stored in an ontology which is a specification of a conceptualization [18]. Ontologies can be created using the resource description framework (RDF) [19], RDF Schema (RDFS), the web ontology language OWL [1] or other standards. Based on these standards web services can be augmented to enable an automatic publication, discovery, interoperability and access. Web services are loosely coupled components working on the internet that provide standardized interfaces and communication protocols aiming at efficient and effective service integration and interoperability between business applications. A semantic web service extends the capabilities of a web service by associating semantic concepts to the web service in order to enable better search, selection and integration. Augmented web services based on these semantic web service standards (e.g. OWL-S [2] or WSMO [3]) contain parameters for describing the functional behavior of a web service including input and output parameter, preconditions and effects, description of the quality of a service, the execution and content of the service, etc. Most of these concepts can be applied to business processes as well. Currently four different standards for semantic web services are emerging: OWL-S, WSMO, SWSF and METEOR-S. In the following these standards will be explained and compared briefly. OWL-S The semantic markup for web services (OWL-S [2], antecessor was DAML-S [20] which is now standardized by W3C) is based on the Web Ontology Language OWL and supplies web service providers with a core set of markup language constructs for describing the properties and capabilities of their web services in an unambiguous, computer-interpretable form. It describes a service in three different ways: Service- Profile ( What does the service provide? ), ServiceModel ( How is the service used? ) and ServiceGrounding ( How does one interact with it? ). To give a detailed perspective on how to interact with a service, it can be viewed as a process. OWL-S distinguishes three kinds of processes: Atomic processes, Simple processes and Composite Processes. Atomic processes are directly callable and correspond to the actions a service can perform by engaging it in a single interaction; composite processes

4 correspond to actions that require multi-step protocols and/or multiple server actions; finally, simple processes are not callable and not associated with a grounding and only provide an abstraction mechanism to enable multiple views of the same process. WSMO Based on the Web Service Modeling Framework (WSMF) [15], the Web Service Modeling Ontology (WSMO) [3] is a formal ontology (based on miscellaneous languages) that consists of four different main elements for describing semantic web services: - Ontologies that provide the terminology used by other elements - Goals that state the intentions that should be solved by web services - Web Services which describe various aspects of a service - Mediators to resolve interoperability problems. SWSF Another W3C submission beside OWL-S and WSMO is the Semantic Web Services Framework (SWSF) [5] which represents an attempt to extend the work of OWL-S and consists of the language SWSL as an underlying basis and the ontology SWSO above. The Semantic Web Service Language and Ontology can be distinguished in two parts: one based on rule languages and logic programming and the other one based on first-order logic enhancing the description logic used in OWL DL. METEOR-S The last proposal for a semantic web service standard is the METEOR-S project [6] of the LSDIS-lab, University of Georgia, collaborating with IBM. METEOR-S is the abbreviation of Managing End-To-End OpeRations for Semantic Web Services and focuses the process itself. The main point of METEOR-S is the use of semantic in the complete lifecycle of semantic web processes to represent complex interactions between semantic web services. A web process is a next generation workflow technology to facilitate the interaction of organizations with markets, competitors, suppliers, customers, etc. supporting enterprise-level and core business activities [22]. The tasks hereby are applying semantics in annotation, quality of service, discovery, composition and execution. To represent semantics in the different stages of a service the project defined four different kinds of semantics: data semantics, functional semantics, Quality-of-Service semantics and execution semantics. Comparison OWL-S provides the distinction between atomic processes, simple and composite processes. This separation is necessary to enable more complex service interactions. The differentiation between ServiceProfile, ServiceModel and ServiceGrounding makes it easier to describe a service completely. The mediator concept of WSMO is very important to solve interoperability issues. The distinction between web services and the underlying ontology is another advantage of WSMO. SWSF extends the OWL-S standard, enables rule languages and extends the description logic used in OWL to a first-order logic which makes it easier to describe concepts and their rela-

5 tionships. METEOR-S separates different ontologies: the data semantics for the definition of the vocabulary, the functional semantics to describe the capabilities of web services, the execution semantics to represent the flow of services and the QoS semantics which explains the quality a service offers. This separation makes it easier to create and modify these ontologies. We will come back to the advantages and concepts of the different standards presenting our own approach. 3 Overall Approach This section describes our overall approach how business or reference process models using UML2.0 activity diagrams can be annotated using the concepts of semantic web services. In particular, we explain how an automated synthesis of processes can be achieved using matrices and algorithms working on the annotation. 3.1 Ontology-based Modeling of Business Processes The modeling of business processes concentrates on the functional view whereas web services are focused on the technical view, i.e. different levels of abstraction. Nevertheless, both services and processes can be described with functional parameters. The semantic web service standards provide descriptions what the service does and how it interacts with others. These descriptions can be applied to business processes and used to describe their choreography. Each semantic web service standard described in section 2 has advantages when being used to annotate business processes. We are interested in a general approach for the automated synthesis of reference processes to be customized and combined to obtain value-added business processes. In order to obtain a realization of the system we decided to build on existing standards. Therefore, we try to combine the concepts and advantages of standards presented in section 2 in our approach, hence being able to switch after it is foreseeable which one is accepted most. OWL-S provides the distinction between atomic processes, simple and composite processes. Business processes as modelled e.g. in an UML2 activity diagram have different levels of detail. A general manager is only interested in the high level process description whereas a project manager needs a more detailed view. Composing several atomic processes to a composite one enables this requirement. Thus we will keep the distinguished views provided by OWL-S. WSMO enables the definition of goals and several tools are already using the web service descriptions and goal description to create a plan for achieving this goal (e.g. IRS-III [25]). These tools and algorithms might be useful for an improvement of our work, but are currently limited to semantic web services in WSMO. Similar to OWL-S SWSF describes each action with four different parameters: input, output, preconditions and effects (in short: IOPEs). This description and abbreviation will be used to describe business processes which will be modelled through actions in UML2.0 activity diagrams. METEOR-S separates different ontologies:

6 - Data semantics: Enables web services to communicate with the same vocabulary. The Data semantics defines the data used in input and output messages of a web service. - Functional semantics: The (semi-)formal description of capabilities of web services is done by annotation operations of web services as well as the provision of precondition and effects. - Execution semantics: This represents the flow of all services in a process (or operations in a service) to analyze, validate and execute process models. - QoS semantics: The Quality-of-Service represents qualitative (e.g. security, transactions, etc.) and quantitative (cost, time, etc.) measures of a web process Our approach will use the data semantics for the specification of the underlying ontology, the functional semantics for the description of each process and the execution semantics to specify how the processes are related to each other. We adopt and merge the advantages of all standards and build the following architecture of ontologies (see Figure 1). Figure 1: Architecture of ontologies Our business process model is defined as follows: BPM = (D, F, E), whereas D is the data semantics, F are all processes (containing the functional semantics) and relations, and E describes the execution semantics. The ontology used to define all concepts of the company and corresponding business processes will be called data semantics and provides a basis for the modeling of processes. It is modelled using OWL and defines all concepts and their relationships that describe the enterprise, the departments and their tasks which are required to annotate business processes. Additionally, (global) variables can be defined which can be used in preconditions and effects for describing the change to a global state, e.g. changes in a database, etc. Each process is described in the functional semantics using IOPEs. Therefore, the concepts defined in the data semantics are used for the input and output section. The variables can be tested for specific values (e.g. true or false) in the preconditions and new values can be assigned in the effects. The functional semantics are currently modelled in a pure version not using any of the available semantic web service lan-

7 guages in order to be able to adapt to the one that will be establishing most. The formal description for the diagram itself would be F = ( 0 i n P i ) ( 0 i,j n Con Pi,Pj ) where n is the number of processes, P i is a process and Con Pi,Pj is a connection between the two processes P i and P j. A connection Con Pi,Pj describes that after process P i has been executed, process P j follows. Each process P i includes the functional semantics for this process, P i = (I i, O i, P i, E i ). The inputs I i and outputs O i are a union of concepts defined in the data semantics (( k 0 k < #inputs: I i k D) ( k 0 k < #outputs: O i k D)). P i describes all preconditions of the process and every precondition is a Boolean expression evaluating variables (tests like int.hasvalue(int):boolean, String.hasValue(String):Boolean, int.lessthan(int):boolean or int.greaterthan(int):boolean). In the effects E i variables that are declared in the data semantics get assigned (variable.setvalue(newvalue)). To improve the automatic synthesis it is possible to define the necessary chronology of some processes. Following the METEOR-S approach this is captured in the execution semantics E. Possible statements for two processes P 1 and P 2 are: - P 1 next P 2 (P 1 P 2, means P 2 follows directly on P 1 ), - P 1 eventually P 2 (P 1 P 2, sometime after P 1, P 2 will be executed), - P 1 previous P 2 (P 2 P 1 ), - P 1 before P 2 (P 2 P 1 ), - P 1 parallel P 2 (P 1 P 2, means P 1 and P 2 are in two parallel threads). 3.2 Synthesis of Ontology-based Business Processes Having created all business or reference sub-processes, i.e. modelled the data semantics, functional and execution semantics, a synthesis task can be started to calculate how these processes might fit together. Therefore, the functional semantics of each process is compared with all other processes in a reasoner. Based on this information a synthesis operation is started. We have developed two different synthesis operations which will be described in section The following section describes how the synthesis matrix and identity matrix are built Synthesis Matrix and Identity Matrix The synthesis and identity matrix are responsible to store the values that are computed in the inference engine. The inference engine has loaded the data semantics and answers queries whether parameters of two processes fit together. It tests the congruence of the outputs of the first process with the inputs of the second whether they are equal or the inputs are a subset of the outputs. Furthermore they are checking whether the preconditions of the second process are satisfied with the effects of the first. The more these tests are successful, the higher is the value in the synthesis matrix. The synthesis matrix synmat is a n n matrix (whereas n is the number of processes in the business model) with i, j n:

8 0, if (P i = P j ) or nomatch(p i,p j ), 2, if ( ϕ ( θ θ < #inputs j, ϕ < #outputs i : I j θ O i ϕ )) 4, if ( ϕ ( θ θ < #inputs j, ϕ < #outputs i : I j θ = O i ϕ )) 5, if α ( β β < #effects j, α < #preconditions i : E j β P i α )) synmat Pi,Pj = 6, if α ( β β < #effects j, α < #preconditions i : E j β = P i α )) 7, if ( ϕ ( θ analog. to 6 : I j θ O i ϕ ) α ( β: E j β P i α )), 8, if ( ϕ ( θ : I j θ O i ϕ ) α ( β: E j β = P i α )), 9, if ( ϕ ( θ : I j θ = O i ϕ ) α ( β: E j β P i α )), 10, if ( ϕ ( θ : I j θ = O i ϕ ) α ( β: E j β = P i α )), whereas = defines that the input and output are equal, describes that the output is a subclass of the input or it can be computed in an inference engine that both concepts are equal, = means that the preconditions can be fulfilled regarding the effects and constrains this to the usage of inference engines (the effects are a sub-concept or equal to another concept that is checked in the preconditions of the other process). We have chosen these numbers for simplification of the implementation: every entry in the matrix bigger than six can be created by adding the smaller numbers and is unambiguously factorable. In the example in Figure 2a the outputs and effects of process P 1 are equal to the inputs and preconditions of process P 2 (value of 10). The outputs and effects of process P 2 are sufficient (but not equal) for the inputs and preconditions of process P 3 (value: 7). Additionally, the outputs of P 3 are similar to the inputs of P 2 (value of 2) and the effects to the preconditions of P 1 (value 5). Taking the matrix as a directed and weighted graph, the graph would be as in Figure 2b. Figure 2: A synthesis matrix and the corresponding graph Similar to the synthesis matrix the identity matrix idenmat contains the results of the comparison of two processes, but this time the equivalence of them is tested. If parts or all inputs are the same (alternatively outputs, preconditions or effects) there is a value bigger 0 (maximum 16), otherwise with a value of 0 the two processes are not parallelizable or the process was compared with itself. 0, if (P i = P j ) or notidentical(p i,p j ), 1, if ( θ1 ( θ2 θ1 < #inputs i, θ2 < #inputs j : I j θ2 I i θ1 )) 2, if ( ϕ1 ( ϕ2 ϕ1 < #outputs i, ϕ2 < #outputs j : O j ϕ2 O i ϕ1 )) idenmat Pi,Pj = 4, if α1 ( α2 α1 < #preconditions j, α2 < #preconditions i : usedvariables(p j α1 ) = usedvariables(p i α2 )) 8, if β1 ( β2 β1 < #effects j, β2 < #effects j : usedvariables(e j β1 ) = usedvariables(e i β2 ))

9 3.2.2 Modified Prim and RandomWalk Having created both matrices these can be interpreted as a directed and weighted graph G = (N, E). The nodes N are the processes and there are edges if the entry in the synthesis matrix between the two processes is bigger than 0 (cf. Figure 2b). Using matrices turns the synthesis problem into a graph-based problem and in this area lots of algorithms are available that can be easily adapted. The synthesis is executed bottom-up: The matrices give information about how two processes fit together and the information about all combinations is used to compute the whole diagram. Therefore, two algorithms were developed: the Modified Prim- and the RandomWalk-algorithm. Modified Prim is an adapted Prim- (or Dijkstra-) algorithm and creates a graph choosing one edge randomly first. Then the edge with the highest value will be added to the solution if it fulfills all given constraints. This is repeated until all processes have been visited. The algorithm in pseudo-code for Modified Prim can be seen in Figure 3. After visiting each process, the operation is terminated and the solution is evaluated. This algorithm is executed with each existing edge as the start edge and after all solutions are computed, the one with the best rating is returned to the user. Select startedge randomly; visitedprocesses = List consisting of startedge.firstprocess & startedge.secondprocess; visitededges = List with single element startedge; while not all processes have been visited do while not edgecorrect do nextedge = getnextedge with highest value not in visitededges; test whether nextedge is a correct edge: it creates no circle there is at least one process in each parallel thread two processes in a parallel thread can be parallized according to the identity matrix the constraints of the execution semantics are fulfilled end while if (edgecorrect) if nextedge has not already been visited add nextedge to visitededges; if firstprocess of nextedge has not already been visited add firstprocess to visitedprocesses; end if if secondprocess of nextedge has not already been visited add secondprocess to visitedprocesses; end if end if end if end while solution = new Solution with visitededges Figure 3: Algorithm of Modified Prim The synthesis algorithm RandomWalk operates on basis of an existing solution (e.g. generated via Modified Prim). RandomWalk was primarily invented to optimize mathematical problems and tries to converge to a maximum by changing the solution iterative. Every solution is rated (using the value of the applied edges) and computed whether there exist solutions with a better rating. Therefore, in our case, edges are removed by accident and others (with a better value) are added. After a period of time the operation is stopped and the so far best solution is returned.

10 4 Case Study To test the ontology-based modeling and synthesis of business processes and compare the different synthesis operations, a prototypical tool was developed [7]. It offers the modeling of a semantically-enriched UML2 activity diagram and testing the synthesis operations explained above. As a case study we will model a business trip. Assuming a business trip to a conference is planed, the following tasks might be necessary: first, plan the business trip including details where the city of the conference is located and which dates for the flights should be scheduled. Afterwards one will search for a cheap flight and book this flight, look for and book a hotel and hire a car. Finally one will make the business trip. This process might be modeled as a reference process and only needs to be customized to a company (e.g. only specific hotels allowed, no flights on Sunday, etc.). <sembpm:object rdf:id= airport /> <sembpm:object rdf:id= airport_of_departure > <rdfs:subclassof rdf:resource= #airport /> <owl:disjointwith rdf:resource= #airport_of_arrival /> </sembpm:object> <sembpm:globalvariable rdf:id= flight_booked /> Figure 4: Data semantics for the process book flight Figure 4 shows a part of the data semantics for one of these processes ( book flight ) and Figure 5 an extract of the functional semantics (written in OWL-S). Having modeled all processes (without transitions!), annotated them with semantics and started the synthesis, one gets the result of Figure 6. Both algorithms achieve the same result and the solution the user might have probably expected. If one of the modeled sub-processes is not required in a company or has additional constraints, one can delete or modify this sub-process and simply start a new synthesis to get the new optimal combination (according to the functional semantics). <sembpm:process rdf:id= book_flight > <owls:hasinput> <owls:input rdf:id= &datasem;#airport_of_departure /> </owls:hasinput> <owls:hasinput> <owls:input rdf:id= &datasem;#airport_of_arrival /> </owls:hasinput> [ ] </sembpm:process> Figure 5: Extract of the functional semantics

11 Figure 6: Result of the synthesis 5 Conclusions and Outlook The presented approach is a first step to use annotated business and reference process models to make an automatic synthesis and compute the best orchestration of all modeled processes. This enables a modeling tool to assist humans in modeling and optimizing a model when changes to the corresponding actions have occurred. This has been a time-omitting task and can now be done without any human interaction. Our approach enables the usage of ontologies to describe business processes and to use this information to compute an optimal workflow which is necessary for an automatic adaptation of reference processes and will increase their utilisability and usage. Our approach combines the advantages of the discussed semantic web service standards, but stays adaptable to each of these standards. We implemented a prototypical application which was developed as an eclipse plug-in and provides therefore an opportunity to be connected with other plug-ins, e.g. for automatic code generation, modifying ontologies, etc. Currently, the tool is more IT-oriented and less business-oriented. To enable business rules, an extension to rule languages like SWRL [27], SWSL-Rules or WSML- Rule would be necessary. Our prototype only offers the modeling of UML2 activity diagrams; other diagrams are currently not supported. Especially diagrams to cover organizational or resource aspects will be included in the next version and additional semantics and ontologies created (e.g. for the organizational structure of a company). In the next steps the functional semantics will receive a fundamental grounding in the UML2 meta-model and a comprehensive profile for the usage of ontologies in business process models will be created.

12 Additionally, we are aiming to develop a concrete mechanism to annotate reference processes, whereas we will consider current approaches to model business processes like BPMN [9] and approaches of business process ontologies like BPMO [8]. References 1. Guiness, D., Harmelen, F.: OWL Web Ontology Language - Overview, February Martin, D. et al: OWL-S: Semantic Markup for Web Services, November Lausen, H., et al: Web Service Modeling Ontology (WSMO), June MID Enterprise Software Solutions, Nuremberg, Homepage: 5. Battle, S. et al.: Semantic Web Services Framework (SWSF), September LSDIS, University of Georgia, METEOR-S: Semantic Web Services and Processes, 7. Programming of Distributed Systems Lab, University of Augsburg, Plug-In and documentation available online at 8. Axenath, B., Kindler, E. and Rubin, V. the Aspects of Business Processes: An Open and Formalism Independent Ontology, Paderborn, April BPMI.org, Business Process Modeling Notation (BPMN) Version 1.0, May Jenz & Partner GmbH: BPEdit What is it? available online at METEOR-S Download and Release Page: METEOR-S Process Designer available online at Supply-Chain Council: SCOR available online at IDS Scheer ARIS Platform on Object Management Group (OMG): Unified Modeling Language (UML) Specification. 15. Fensel, D, Bussler, C.: the Web Service Modeling Framework WSMF, In: EC Business Process Management Initiative (BPMI): Business Process Modeling Notation (BPMN), Version 1.0, May LSDIS, University of Georgia, MWSAF: METEOR-S Web Service Annotation Framework, Homepage: Gruber, T. A translation approach to portable ontology specifications, W3C: RDF - Resource Description Framework, The DAML Services Coalition: DAML-S: Semantic Markup for Web Services, debrujin, J., Lausen, H.: Web Service Modeling Language (WSML), June Cardoso, J. and Sheth, A.: Semantic Web Processes, 4 th International Conference on Web Information Systems Engineering (WISE 2003), Rome, December McIlraith, S. and Son, T.: Adapting Golog for composition of semantic web services. In: KR-2002, Toulouse, France, Au, T.C., Kuter, U. and Nau, D.: Web Service Composition with Volatile Information, In: ISWC05, Galway, November Domingue, J., Galizia, S. and Carbal, L.: Choreography in IRS-III Coping with Hetereogeneous Interaction Patterns in Web Services, In: ISWC05, Galway, Greenwood, D., Buhler, P. and Reitbauer, A.: Web Service Discovery and Composition using the Web Service Integration Gateway, in: EEE Horrocks, I. et al.: SWRL A Semantic Web Rule Language Combining OWL and RuleML, W3C Submission, May Koschmider, A. and Oberweis, A.: Ontology based Business Process Description, in: EMOI-INTEROP 05, Porto, June 2005.