Linking Heterogeneous Conceptual Models through a Unifying Modeling Concepts Ontology Janina Fengel Michael Rebstock University of Applied Sciences Darmstadt University of Applied Sciences Darmstadt Hohe Strasse Haardtring 100 D – 64807 Dieburg D – 64295 Darmstadt +49 6151 16 9458 +49 6151 16 8392 janina.fengel@h-da.de michael.rebstock@h-da.de ABSTRACT exist. There are efforts in matching models concentrating on the In this discussion paper we report on our ongoing work in aspect of model language semantics based on migration or applying Semantic-Web technologies for supporting business transformation from one modeling language into another [12; 7; integration. Our method foresees the reengineering of conceptual 10], matching models via their meta models [9] or concentrating models into ontologies for performing domain-semantics oriented on managing models of the same kind [11]. Thereby, the aspect of matching. For linking models in differing modeling languages as heterogeneously used domain language is not addressed, instead well as different model types, we have developed a bridge the model element labels are transferred and retained unchanged ontology. The first results show the feasibility of our approach. for further use. Extending process models with semantic annotations for easing consistent modeling and business-IT align- ment has been suggested, thus turning models into model Categories and Subject Descriptors instances [16; 2]. For assigning element labels, the use of a H.4.0 [Information Systems]: Information Systems Applications separately developed domain ontology has been proposed, similar – General. as in the suggestions for semantic business process management, which rely on such a pre-defined business terminology [8; 4; 15]. General Terms It can assist in the creation of new models and provide the basis Management, Standardization, Languages. for unambiguous element labeling as well as serve for mediating the matching of existing models. However, the creation of a common domain ontology or business terminology to be used as a Keywords standard is usually time-consuming and cost-intensive. Further- Conceptual modeling, semantic heterogeneity, domain semantics, more, comparing a model to the set standard is still labour- modeling languages, bridge ontology intensive work. For easing this workload, we propose to convert existing process, 1. INTRODUCTION data and organizational models into ontologies and provide The need for integrating conceptual models arises at the time of automated support for relating them by means of ontology process optimization, business (re-)engineering or generally in matching techniques. In this, our approach may serve as a business integration. Typical situations are reorganizations or complement to the existing works in process matching as mergers, leading to process and application integration outlined, as it offers a means to semantically integrate models of challenges. Upon integrating the conceptual models describing different kind regarding the domain and modeling language the business operations and the underlying IT-support, together. Additionally, through the semantics-oriented reuse of heterogeneously used natural language for labeling model element the domain knowledge contained in models, over time the labels often hinders meaningful comparison. Furthermore, the collection of linked models may be taken as a skeleton semantic usage of different models in differing modeling languages usually domain or, more specifically, enterprise or business ontology. In prevents automated support in aligning, linking or merging the following we continue with describing our method of models. Nevertheless, models to be integrated need to be converting models into ontologies, followed by presenting the compared regarding the intended meaning of their elements and bridge ontology specifically designed for enabling semantic their structure, whereas structural analysis cannot be performed integration. We conclude with showing the method’s application until successful alignment of the domain language [13]. Thereby, using a small example, closing with a brief discussion and outlook especially naming conflicts hinder model integration [1; 14]. In onto our further work. practice, often differing unrelated non-aligned legacy models 2. ONTOLOGIZING MODELS Permission to make digital or hard copies of all or part of this work for By ontologizing models the business concepts’ semantics are personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that made machine-accessible, independent of the modeling languages copies bear this notice and the full citation on the first page. To copy used. The creation of a model requires knowledge of the domain otherwise, or republish, to post on servers or to redistribute to lists, language for naming the business concepts to be described as well requires prior specific permission and/or a fee. as the modeling language for describing their relations and sorting SBPM 2010, May 31, 2010, Heraklion, Crete, Greece. them. Reversing this process allows us to decompose a model into Copyright 2010 ACM 1-58113-000-0/00/0010…$10.00. 1 the domain semantics separately from the modeling language Input for the development of the UMCO has been drawn from semantics [5]. A model and its elements are split into two separate existing enterprise modeling languages and ontologies. Business ontologies in OWL DL, which together describe the model with process modeling languages provide the means for describing its model type and name and the model elements with their model sequences of activities. They offer the idea of activities, either element types and labels. In the model conversion, all labels are being called activities, tasks, functions or actions, which start and presently taken without further processing, so that not only terms, end with an event and are linked by flows. For the description of but complete expressions are transferred. Often, domain the behavioral aspect of processes, the flows can be tied to logical knowledge in the field of business processing lies in the connectors for making decisions and showing alternative flow combination of objects and the execution of activities, which is paths [12]. In detail the semantics of process modeling languages preserved this way. Basically, the suggested decomposition are not equivalent, so that models cannot be translated directly method abstracts from the statement a model intends to do and without loss of information [10]. However, the fundamental leaves the model as is for further active use. Figure 1 shows the intensions of the concepts are comparable. The same observation meta model of a thus decomposed conceptual model. can be made for models describing static business information. Conceptual data models can be represented as entity-relationship models, UML class models or directly as OWL-ontologies. Thereby, the entities of an ERM or classes of a class model correspond to the classes of an ontology, while the attributes and relationships correspond to the relations or properties in most ontology languages [17]. Furthermore, UML class models can be used for ontology modeling [6; 3]. For our purpose of integrating models, we have defined general modeling concepts and declared them equivalent to corresponding concepts in the various modeling languages. In this, the UMCO further extends the meta model shown in Figure 2. Figure 1. Meta model of a decomposed model Thereby, the domain facts expressed in the natural domain language are separated from the type of element they are connected with. This type information resembles attaching prove- nance information. For doing so, the idea of indexing the domain facts in a manner similar to indexing in librarianship in form of Topic Maps has been adopted. The model ontology on the left side captures the domain knowledge in natural language as owl:Classes and the relations between them as properties Figure 2. Meta model of a decomposed model with relations to with restrictions as needed. This model ontology links to the the UMCO model type ontology. For each modeling language a specific The UMCO is extensible as needed for including further MCO- Modeling Concept Ontology (MCO) has been developed, contain- ontologies. Developing MCOs for process, data and organization ing parts of its meta model. The domain knowledge expressed in models, or any other models, allows connecting knowledge not the logical relations between model elements as the means for only directly for model ontologies originating from the same setting the specific models’ element order is preserved together language space of the modeling languages, but additionally, with the domain facts in the model ontology, not in the MCO. through the unifying modeling concepts, also knowledge from This conversion returns the element labels representing business model ontologies of different modeling languages. For example, concepts as classes, thereby allowing a later ontology extension an EPC is defined compliant with an UML activity model and a with the concepts’ instances. Thus, the principle of conceptual BPMN model, and these concepts are set to be equivalent to the modeling in business is carried forward. UMCO concept called UMCO:Process; EPC_MCO:Function in an EPC, UML_AM_MCO:Action in an UML activity model and 3. THE UMCO AS A BRIDGE BPMN:Task in a BPMN model are set to be equivalent to the In principle, for any type of model an MCO can be developed. In concept called UMCO:Activity. Accordingly, all modeling order to be able to link the ontologies resulting from the conver- concepts found in the various languages can be unified and sion as described, the MCOs enable references between models of related, e.g. linking resources in process models such as the same type. For further enabling also the referencing of models documents or participants to the data models detailing them. of the same kind, but different type as well as also models of different kind, we use the Unifying Modeling Concepts Ontology (UMCO), which we have developed specifically for this purpose. 4. LINKING MODELS Usually, matching models is a major task. To ease this workload, It provides a unifying model concept for each type of modeling we partially automate this step by reengineering the conceptual concept with a similar intention. In this, it represents all relevant domain knowledge contained in existing models as shown and element types for our approach, usable like a meta-meta model. relate the resulting model ontologies semantically for establishing The UMCO and its MCOs serve for linking modeling concepts semantic correspondence between the model ontologies’ without predefining relations beyond part-whole-definitions, as elements. As an implementation for semantic model integration in this is done in the model ontologies. the described manner, we develop the MODI (Model Integration) Framework as an application of our method. Our framework is 2 realized in Java and can be accessed by a web service interface. It initial base is being established by means of automated tools. At consists of a core component, to which tools for mapping and the time of using the resulting mapping ontology, the automati- storage can be variably connected by adapters. Figure 3 shows the cally derived information is evaluated by its users. Even though architecture. the need for manual work is reduced, the mappings found are not always perfect, but may be ambiguous or incorrect [18]. Therefore, our system facilitates user participation by enabling adding, editing and feedback provision for growth and improvement over time. The evolving repository can be queried for semantic correspondences. Thereby, a user may request references for a specific term or directly compare two ontologies. Figure 5 shows the prototypical screen of the results for a comparison of the two example business process models. Figure 3. MODI Framework Architecture The focus of our work is on the realization of matching and establishing domain-semantics based mappings between models. Having performed the model conversions as described above, ontology matching can be performed without merging any of the input model ontologies. As the model ontologies obtained by converting process models do not contain hierarchical or mereologic relations, only element-based techniques return meaningful mappings, best by tokenization and name matching. Figure 5. Prototypical list of suggested semantic For matching converted ERM and class model ontologies, also correspondences structure-based techniques can be used, as here in most cases subsumption and aggregation relations return exploitable With an increasing number of models included, first tendencies ontology structures. Since the domain facts are not transferred as towards commonly used terms can become obvious. An initial instances of their individual model type, it is prevented that terminological domain ontology emerges, consisting of the matchings return mappings between model element types. These various independent model ontologies, which are linked through links are provided without creating workload for the matchers the mappings stored. This emerging ontology can be used at the through the introduction of the MCOs and the UMCO. time of creating new models searching for a suitable element label Furthermore, avoiding such an undesired hierarchical structure as well as for explaining the intended meaning in existing models focuses the matching efforts onto the domain language that need to be compared and related semantically. independently of the original modeling language used. Figure 4 Combining the model ontologies with our modeling concepts shows excerpts from two converted business process models from ontologies allows for searching for specific model types., e.g., the travel domain as an example. Each model depicts the booking searching for all EPCs available, as well as for models of all types of services. The source model “Travel Reservation” is an EPC, of a certain kind, e.g., such as process models either being EPC or while the target model “Travel Booking” is a UML Activity UML Activity Models, through utilizing the corresponding Model. They both depict the process of booking travel services, unifying concept, here UMCO:Process, for detailing the query. however, the domain language differs. Alternatively, searches for UMCO:Activity return all business operation steps. With the method described, not only models of the same kind can be matched and related. Instead, linking different models is possible as well with the help of the various MCOs und the UMCO. 5. DISCUSSION AND CONCLUSION Here we proposed an approach based on applying ontology engineering techniques for achieving semantic integration of conceptual models in the business domain. A method for reusing existing conceptual models and relating the business knowledge Figure 4. Excerpts from two linked business process models contained without huge manual efforts is shown. We have created The matching works could be performed successively as needed. a way for reengineering such non-ontological resources for mean- Results became available from the beginning, especially after ingful relating and linking with the help of supporting ontologies having included lexical background information from WordNet. especially created for this purpose. The related models can be In our framework, all mappings found as a result of matchings are analyzed and compared regarding the intended meaning of their stored in a repository as semantic correspondences. Thus, an elements. The automatically produced results provide a basic 3 lightweight domain ontology without initial manual preparation [6] Gašević, D., Djurić, D., Devedžić, V., and Damjanović, V., and creation efforts. By including user input, a possibility for 2004. Converting UML to OWL ontologies. In Proceedings overcoming the shortcomings of automated knowledge computing of the thirteenth International World Wide Web Conference. is presented, as human support is included for assessing and ACM Press, New York, 488–89. improving the quality of the mappings found by way of [7] Gehlert, A., 2007. Migration fachkonzeptueller Modelle. automated matching. The resulting mapping collection provides Logos-Verl., Berlin. support for the clarification of uncertainties and allows for semantically integrating models of any type. Our system works as [8] Hepp, M., Leymann, F., Domingue, J., Wahler, A., and a mediating medium and helps providing the grounds for Fensel, D., 2005. Semantic Business Process Management: A concentrating on the actual questions of process integration and Vision Towards Using Semantic Web Services for Business activity sequencing. Process Management. In Proceedings of the IEEE International Conference on e-Business Engineering ICEBE The system presented here has been implemented as a 2005 (Beijing, China, Oct. 2005). IEEE Computer Society, prototypical solution for the method developed and is being Los Alamitos, 535–40. evaluated. Its application has proven the conceptual strength and [9] Kappel, G., Kramler, G., Kapsammer, E., Reiter, T., practical relevance. Still, a number of aspects remain to be Retschitzegger, W., and Schwinger, W., 2005. ModelCVS - researched. Since the system is based on coupling existing tools A Semantic Infrastructure for Model-based Tool Integration. for ontology matching and storing via adapters, evaluation Technical Report. University of Vienna. concerning their efficiency and performance is of interest. The research concerning the comparison and combination of mapping [10] Kensche, D., Quix, C., Chatti, M. A., and Jarke, M., 2007. tools has to be concluded. Furthermore, as also is the case with GeRoMe: A Generic Role Based Metamodel for Model social software, our system needs a critical mass of users in order Management. Journal on Data Semantics VIII, LNCS 4380, to be useful. Hence, it needs to be proven that the method and Springer, Berlin, 82–117. with this also our framework offers benefits. Overall, with our [11] Melnik, S. Rahm, E., and Bernstein, P. A., 2003. Rondo: A approach of semantic model integration we hope to have shown Programming Platform for Generic Model Management. In how the usage of Semantic-Web technologies may support Proceedings of the ACM SIGMOD International Conference business modeling. on Management of Data (San Diego, California, June, 2003). ACM Press, New York, 193–204. 6. ACKNOWLEDGMENTS [12] Murzek, M. and Kramler, G., 2007. The Model Morphing Our work is supported by the German Federal Ministry of Approach – Horizontal Transformations between Business Education and Research (BMBF) under grant no. 1728X07 Process Models. In Proceedings of the 6th International MODI. Conference on Perspectives in Business Information Research - BIR'2007. Vol. D‐2007‐13. Tampere, Finland, 7. REFERENCES Department of Computer Sciences, 88–103. [1] Becker, J., Rosemann, M., and Schütte, R. 1996. [13] Simon, C. and Mendling, J., 2007. Integration of Conceptual Prozeßintegration zwischen Industrie- und Handelsunter- Process Models by the Example of Event-driven Process nehmen - eine inhaltlich-funktionale und methodische Chains. In Proceedings of Wirtschaftsinformatik WI 2007 Analyse. Wirtschaftsinformatik 39(3), 309–316. (Karlsruhe, Germany, March 2007). Vol. 1. Univ.-Verl., [2] Belecheanu, R., Cabral, L., Domingue, J., Gaaloul, W., Karlsruhe, 677–94. Hepp, M., Filipowska, A., Kaczmarek, M. et al. 2007. [14] Thomas, O. and Fellmann, M. 2006. Semantische Integration Business Process Ontology Framework. Deliverable 1.1. von Ontologien und Ereignisgesteuerten Prozessketten. In SUPER Project. Proceedings of EPK 2006 - Geschäftsprozessmanagement [3] Brockmans, S., Volz, R., Eberhart, A., and Löffler, P. 2004. mit Ereignisgesteuerten Prozessketten (Vienna, Austria, Nov. Visual Modeling of OWL DL Ontologies Using UML. In 2006). Vol. 224. CEUR-WS.org, 7–23. Proceedings of the Third International Semantic Web [15] Thomas, O. and Fellmann, M. 2007. Semantic Business Conference ISWC 2004 (Hiroshima, Japan, Nov. 2004). Process Management: Ontology-Based Process Modeling Springer, Berlin, 198–213. Using Event-Driven Process Chains. IBIS Interoperability in [4] Brockmans, S., Ehrig, M., Koschmider, A., Oberweis, A., Business Information Systems 2(1), 29–44. and Studer, R., 2006. Semantic Alignment of Business [16] Thomas, O. and Fellmann, M., 2009. Semantische Prozess- Processes. In Proceedings of the Eighth International modellierung – Konzeption und informationstechnische Conference on Enterprise Information Systems ICEIS 2006 Unterstützung einer ontologiebasierten Repräsentation von (Paphos, Cyprus, May 2006). INSTICC, Setúbal, 197–203. Geschäftsprozessen. Wirtschaftsinformatik 51(6), 506–518. [5] Fengel, J. and Rebstock, M., 2009. Model-Based Domain [17] Uschold, M. and Gruninger, M., 2004. Ontologies and Ontology Engineering. In Proceedings of the 4th Semantics for Seamless Connectivity. ACM SIGMOD International Workshop on Semantic Business Process Record 33(4), 58–64. Management SBPM2009 (Heraklion, Crete, June 2009). [18] Zhdanova, A J. de Bruijn, K. Zimmermann, and F. Scharffe, ACM Press, New York. 2004. Ontology Alignment Solution v2.0. Deliverable D1.4 V2.0, Esperonto project. 4