Assessing Information Loss in EPC to BPMN Business Process Model Transformation

Assessing Information Loss in EPC to BPMN Business Process Model Transformation Olga Levina Department of Systems Analysis and IT Berlin Institute of Technology, Germany olga.levina@sysedv.tu-berlin.de Abstract Much research work has been discussing the transformation between different process modeling languages. The focus was often put on model syntax rather than on the integrity of process content. Here an established approach for evaluating content is used to assess the content change that results due to process model transformation from EPC to BPMN business process representation. The results show that induced change is not significant and provides insights into EPC and BPMN language structures. Contribution of this paper is thus a generic strategy for content measurement in business process models as well as theoretical view on business process modeling notations. The findings have implications for analysts and modelers in that they provide guidance on language choice and reduce uncertainty of model transformation. Key words Process modeling, content, empirical study, EPC, BPMN I. INTRODUCTION Business process models are components of efficient and structured management of organizations and enterprises. There are different incentives to model business process in enterprises being e.g. business process re-engineering, documentation or visualization. Nonetheless, creation and maintenance of business process models requires a significant effort by the enterprise or organization in terms of personal resources, time and costs [14]. Therefore, process models constitute a valuable asset for the enterprise that needs to be preserved and maintained. Numerous modeling notations were designed to capture different process aspects for process documentation and visualization. One of the popular modeling notations for business process modeling is the EPC (event-driven process chain). Starting in the 1990ies various enterprises created numerous business process models using this notation that are still being used as basis for process execution and management. More recently, due to the movement towards international collaboration between enterprises and the need for a common modeling language, Business Process Modeling Notation (BPMN) has been developed and standardized. Being an international standard, BPMN is gaining importance in the context of international collaboration between enterprises. Hence, enterprises that have put the effort to model their processes using EPC and are supposed to change to BPMN due to the external or internal incentives need to address the question what effects will the process transformation to BPMN have. Thus, this is the research question of this paper: Does content change when a business process model designed in EPC is remodeled in BPMN? A subsequent question that arises is: which of the two process modeling notations provides the more efficient process representation? This paper addresses these questions by analyzing changes in the content of the model that are induced by structural transformation between languages using theoretical approach. Applying the definition by [5] and [12], is considered here as a technical unit. Its content can be quantitatively perceived based on the ideas of transmission allowing the quantification of the content of a message as well as the definition of data or elements that are redundant. Hartley [5] considered communication as an exchange of symbols that belong to a specified set. Shannon enriched this approach by introducing the probabilistic view on. He defined as the probability of certain sequences of elements from a defined set. In this paper the set of symbols is the set of elements of the modeling notation under consideration, i.e., EPC and BPMN. The message is the business process model. Here an empirical analysis of 30 business process models that were modeled using EPC and then transformed to BPMN (version 1.2) are analyzed. The contribution of this paper is the abstraction from the process semantic content of a process model enabling a generic approach to investigation of al content determination. Thus, this paper does not address consistency and semantic aspects of business process model transformation. Empirical results presented here show that transferring business process descriptions from EPCnotation in BPMN does not significantly affect its content, iff the transformation is performed according to the original model, i.e. the aspects covered (mapped) in the original model are re-modeled using the new notation. Furthermore, analysis results indicate that EPC is providing a more efficient alphabet (i.e., set of modeling elements) than BPMN, whilst BPMN

provides a more efficient use (i.e., modeling guidelines) of the elements for content representation. These findings are interesting for enterprises that need to transform existing process models as well as for process analysts exploring or designing a supporting business application system. Furthermore, these results can support the development of modeling guidelines regarding their efficiency. Reminder of the paper is structured as follows: adjacent to the related work overview, the subsequent section introduces the research method and aspects tested in this research. Then, BPMN, EPC as well as measurement are introduced followed by the research design description. Discussion and outlook on further research finish the paper. II. RELATED WORK In the resent business process model research the focus is increasingly laid on EPC and BPMN transformation, as many enterprise process models are stored in the EPC notation. [14] provides significant insights on the transformation technique for EPC to BPMN- model transformation. He provides transformational guidelines focusing on preservation of semantic content including the problem of mapping EPC events to BPMN. These findings were also used here to validate the analyzed BPMN models originating in EPC. As this paper investigates content of processes presented in different modeling notations, related work on modeling method evaluation is also reviewed here. [13] report techniques for evaluating modeling methods and differentiate with one of them being theoretical and conceptual investigation. This technique is applied in the presented research as it has developed means of evaluating modeling methods without the need of access to empirical data by carrying out analytical studies in well-defined and narrow subject areas based on established terms of reference such as e.g. meta-models [4] and ontology [16]. Metrics analysis [1], [6], [8], [15] uses meta-model based metrics to evaluate the structural properties of a modeling method in order to make statements about its complexity. The premise in this approach is that the presence of more modeling elements increase cognitive load for the modeler [2]. In the area of BPMN-related research [7] use count metrics by [8] to evaluate method complexity of UML versus BPMN version 1.2. They discovered the correlation between complexity and the absolute number of elements in the language and thus that rules and constrains of BPMN are a significant source of complexity [7]. Furthermore, [17] investigate what alphabetical subsets of BPMN are used by academic, consulting and general use of the language. They analyzed 120 BPMN models using mathematical and statistical techniques. Their findings show that less than 20% of BPMN vocabulary is regularly used. III. INTRODUCTION TO EPC AND BPMN Event-driven Process Chain (EPC) notation was originally developed by [9], [10] and has been modified in numerous projects [11], [14]. The original set of EPC elements consist of: event, function, connectors and control flow. This basic representation form has been extended and modified by additional elements representing organizational units or positions as well as data or al objects as the language is not formally normed or standardized. BPMN is a notation standard developed by an industry consortium and is based on UML, ebxml as well as EPC among other notations [17]. BPMN constructs are specified and documented and grouped into four basic categories: flow objects (events, tasks and gateways), connecting objects (e.g., sequence and message flows that connect flow objects), swimlanes (pools and lanes that group activities to organizational units and roles) and artifacts (e.g. data or comments). BPMN (version 1.2) can be used to describe processes at different levels of abstraction, i.e. from value chain to transactional processes [7]. BPMN in version 2.0 was released in January 2011. IV. INTRODUCTION TO INFORMATION MEASUREMENT IN BUSINESS PROCESS CONTEXT To approach measuring, Shannon s theory has been used here as a wide-spread method for quantification using alphabetbased message analysis. Hartley [5] laid the ground for quantification assuming that communication between several parties involves a group of symbols, here also called the alphabet, that have a specific meanings for the communication parties. In the context of business process modeling this alphabet is the set of defined elements of the considered notation. Shannon [12] enriched this interpretation using the probabilistic approach. Thus, content of an element of the alphabet depends on its probability of occurrence. In a set with n independent symbols the probability of occurrence of an element e, p(e), is 1/n. Shannon defined the content I e of an element e: I e = log 2 1/p(e) (1) The average content of a message is also called entropy H and is defined by Shannon [12] as: H = - n i=1 p i* log2p i (2) Being the most widespread and proven approach for quantification, this approach is mapped on the business process modeling context as a process model is used to transport. A message in the context of business process modeling is the business process model under analysis. The size of the model is defined here as the number N of elements present in the model for process description. Combining the equations (1) and (2) content of a business process model can be calculated according to the modeling alphabet, i.e. notation elements

used in the model, as well as to their frequency of occurrence. Thus, the content of a process model is defined in terms of occurrence of alphabet elements rather than under semantic aspects. V. RESEARCH DESIGN A. Experiment planning and execution To approach the research questions on content change as well as content in the different notations experimental design was developed focusing on process flow representation. A set of 30 process models that are modeled in the EPC-notation and were also converted in BPMN with the goal to preserve process, i.e. mapping the EPC representation to the BPMN model, was gathered from practice and research publications as well as industry projects in the area of business process management. These models originate in the context of model transformation due to standardization efforts of modeling notation or international collaboration between enterprises involving already modeled processes. The only requirement for the model analysis was that the process had to be already modeled in EPC and in BPMN accordingly, thus implying that the models were sufficiently similar in their content. EPC and BPMN as well as the transformation process have been exhaustively described by [14]. The author has also defined some general and semantic rules for the model transformation. Even though the process models were designed using a common notation, they can still diverse in their quality, i.e. concerning their structure, understandability, granularity or completeness. Thus, and because EPC is not a standardized notation, modeling requirements were established according to [14] for both notations. Following validation of the model as well as the number and frequency of elements used were elevated and al content of the model was calculated according to equation (2). These results were analyzed using methods of inferential statistics. Guiding assumption of the analysis was that the models were formally correct, therefore they have been beforehand evaluated for formal correctness according to specified modeling guidelines based on BPMN and EPC specifications as well as to the requirements mentioned above. Furthermore, the models were analyzed for content-related equivalence, i.e., the BPMN model had to include the aspects that were expressed in the EPC model. Following aspects were in the focus of analysis: A1: Change of content due to model transformation from the EPC- to BPM-Notation. A2: Use of the set of elements n to represent the business process in EPC and BPMN. A3: Information content per notation element, i.e. element of the alphabet. A4: Information content per modeled element in the process diagram. To assess these aspects the t-test as well as correlation analysis were used. Results of the statistical analysis are presented in the following sections. B. Assessing the Change To assess A1, al content was calculated for the EPC- as well as for corresponding BPMN- models according to the equation (2). The two groups were tested regarding significance of their content difference using the t-test with the level of significance of α = 5%. The results in table I (including standards deviation) show that BPMN models have slightly higher average content than the original EPC models. Nevertheless, this difference is not significant. A2 focuses on the size of the alphabet, i.e., number of defined notation elements n. Table 1 shows that the EPC alphabet is significantly smaller than the BPMN alphabet. On the other hand, the number of modeled elements N in an EPC-model N(EPC) is significantly (with α = 1%) higher than the number of modeled elements in the BPMN-model of the same process N(BPMN): N(EPC) >> N(BPMN). Furthermore, the question arises, whether the content per EPC-element of the alphabet is different from the one per a BPMN-element of the alphabet for the same process. The comparison results show that the content per element of the alphabet is significantly higher for the EPC alphabet comparing to the BPMN alphabet. A4 addresses the effect of the number of elements modeled in a process representation on the content per modeled element. Here the content per element present in a BPMN-model is significantly (with α = 1%) higher than the content per modeled element in the original EPC-model. Table I provides an overview of the abovementioned t-test results including the value of significance of the occurring differences. These results are further discussed in the following sections. EPC BPMN p content of a process model 2.16 (0.26) 2.21 (0.30) 0.436 number 98.7 (74.14) 56.70 (32.41) 0.007 of modeled elements (avg. N) size of 5.87 (1.28) 7.67 (1.18) 1*10{}^{-5} alphabet (avg. n) 0.38 0.29 1*10^{-6} content per alphabet element 0.03 0.05 0.007 content per element in the model

TABLE I. T-TEST RESULTS Additionally, relation of content to the number of modeled elements was explored for EPC and accordant BPMN models. Only the BPMN-models showed a negative correlation on the level of significance of α = 5% between the model entropy H(BPMN), i.e., its content, and the number of modeled elements for process representation N(BPMN). Information content of EPC-models does not show any accordant significant correlation. A negative correlation is also observed between the number of used EPC-alphabet elements n and the content per element used. Furthermore, a negative correlation on the level of significance of α = 1% between the sum of elements in the original EPCprocess model N(EPC) and content per modeled element in BPMN and EPC as well as content per element of alphabet in the BPMN model, is observed. As expected, the number of modeled elements of the BPMN- model N(BPMN) is strongly (on the level of significance of α = 1%) positively correlated with the number of modeled elements in the EPC- model N(EPC). A positive correlation on the level of α = 1% of the content per alphabetic element of the BPMN-model and per modeled element in the BPMN-model as well as content per EPC-modeled element and content per alphabetic element used in the BPMN model can be observed. On the level of significance of α = 5% a correlation between the content per alphabetic element of the EPC-model and content per modeled element in the BPMN-model is observed. Additionally, correlation between H = H(BPMN)- H(EPC) and the size of the alphabet n used in the EPC-model was explored. A positive correlation between the two quantities was stated on the level of significance of α =5%. C. Interpretation and Discussion Approaching the question of measuring change in transformed business process models resulted in two different outcome categories. First of all, significance of the impact of model transformation from EPC- to BPMN-notation has been measured. Relation between original notation alphabet and its impact on the notation alphabet that was used for transformation were explored. The quantitative approach to led to first insights on content transported by a notation element according to its occurrence probability. This approach is not limited to EPC- and BPMN process representation. Thus, it can be used to define al content (change) in other business process modeling notations. The effect of notation change has been examined using t- test. A1 addressed the loss of model transformation measured by the entropy metric. The entropy of BPMN models increased. This fact is interpreted by Gell-Mann [3] as loss. Hence, the transformation of the business process model in EPC to BPMN leads to significant loss. Nevertheless, this loss showed to be not significant in the presented experiment. A2 explored the range of the alphabet that was used to represent the process content. Analysis showed that EPC uses a significantly smaller alphabet than BPMN implying a more efficient alphabet for business process representation. On the other hand, BPMN has a higher range of elements, i.e., larger alphabet, than an extended EPC. This variety results in a less efficient alphabet. Nevertheless, further analysis of the number of elements used or present in the process model, i.e., its size, showed that an EPC-model contains a significantly higher number of elements than BPMN, suggesting that EPC modeling guidelines render the efficient alphabet less efficient. On the other hand, BPMN, having a less efficient alphabet, is supported by modeling guidelines that allow its efficient use. Correlation analysis was used here to explore the influence of the alphabet or model size of the original model on its content. Thus, large EPC-models lead not only to reduced content per modeled element in both notations, but also to the loss of content per alphabet element in the BPMNmodel. Furthermore, diversity in use of the EPC-alphabet results in loss of content per element modeled for process representation. Size of the BPMN- model influences significantly its amount of content H, while the size of the EPC-model does not have any significant impact on its entropy, suggesting that BPMNmodels are more sensitive to changes in the number of modeled elements supporting the abovementioned suggestion that BPMN has more efficient modeling guidelines. VI. SUMMARY AND OUTLOOK In this paper an approach for measuring content of a business process model based on Shannon s theory was presented. This approach was used to analyze change in content for models that are transferred from EPC to BPMN notation. The study allows insights into the content, defined in terms of occurrence probability of alphabetical elements, of a business model as well as measuring the change in the content induced by transformation of this model into a different notation. This view focuses on the model structure disregarding its semantics or consistency aspects. This view is based on the current need of enterprises to address the transformation of their available EPC process models to the standardized BPMN language and the subsequent question whether this transformation impacts content of the model. Presented findings suggest that content difference from model transformation from EPC to BPMN is not significant, iff the transformation considers aspects modeled in the original process representation. Furthermore, EPC alphabet showed higher

content per elements than BPMN, whilst EPC provided less per modeled element, possibly due to its modeling guidelines. Diversity of used EPC-alphabet elements resulted also in decline of content per alphabet element used, thus reflecting the findings by [7]. The here presented simplified view on the calculation of the content of a business process model did not consider the different probabilities of an alphabet element. Thus, further insights on the element frequency distribution are required. To derive reliable probability distributions for EPC- and BPMN alphabet elements, further analysis of numerous models from the transformation context is required. Additionally, exploration of content change on a more abstract level, such as transformation between block- and graph-oriented modeling languages, will be preformed. [17] M. zur Muehlen and J. Recker. How much language is enough? Theoretical and practical use of the business process modeling notation. In 20th International Conference on Advanced Information Systems Engineering, Montpellier, France, 2008. Springer. REFERENCES [1] A. Bajaj and S. Ram. Evaluating completeness of conceptual business process models: A metric based on case studies. Journal of Information Technology Cases and Applications, 1(4), 1999. [2] P. Chandler and J. Sweller. Cognitive load theory and the format of instruction. Cognition and Instruction, 8(4):293 332, 1991. [3] M. Gell-Mann. The Quark and the Jaguar: Adventures in the Simple and the Complex. St. Martin s Griffin, 1995. [4] T. A. Halpin. Metaschemas for er, orm and uml data models: A comparison. Journal of Database Management, 13(2):20 30, 2002. [5] R. V. L. Hartley. Transmission of. Bell System Technical Journal, pages 535 564, 1928. [6] E. Marcos, J. Cervera, and L. Fernandez. Evaluation of data models: A complexity metric. In K. Siau, editor, International Workshop on Evaluation of Modeling Methods in Systems Analysis and Design, volume 4, Heidelberg, Germany, 1999. [7] J. Recker, M. zur Muehlen, K. Siau, J. Erickson, and M. Indulska. Measuring method complexity: Uml versus bpmn. In AMCIS 2009, 2009. [8] M. Rossi and S. Brinkkemper. Complexity metrics for systems development methods and techniques. Information Systems and Computers, 21(2):209 227, 1996. [9] A.-W. Scheer. ARIS- Modeling Methods, Meta-models, Applications (in German). Springer, Berlin, 1998. [10] A.-W. Scheer. ARIS - Business Process Modeling. Springer, 3rd edition, 1999. [11] A.-W. Scheer and W. Jost. ARIS in Usage: Design, Implementation and Optimization of Business Processes (In German). Springer, Berlin, Heidlberg, 2002. [12] C. E. Shannon. A mathematical theory of communication. Bell System Technical Journal, 27(July-October):379 423, 623 656, 1948. [13] K. Siau and M. Rossi. Evaluation techniques for systems analysis and design modelling methods - a review and comparative analysis. Information Systems Journal, 21(3):249 268, 2009. [14] W. Tscheschner. Transformation from epc to bpmn. Business Process Technology, 1(3):7 21, 2006. [15] I. T. P. Vanderfeesten, H. A. Reijers, J. Mendling, W. M. P. van der Aalst, and J. Cardoso. On a quest for good process models: The crossconnectivity metric. In Z. Bellahsene and M. Léonard, editors, CAiSE 2008, pages 480 494, Montpellier, France, 2008. Springer. [16] R. Y. Wang, H. B. Kon, and S. E. Madnick. Data Quality Requirements Analysis and Modeling. MIT Press, 1993.