Reference Model Mining

Die Referenzmodellierung (ausführlich: Referenzinformationsmodellierung) ist ein spezielles Arbeitsgebiet der Informationsmodellierung, in dem Referenzmodelle betrachtet werden. Im Mittelpunkt stehen Fragen der Wiederverwendung von Modellinhalten. Unter Referenzmodellen werden Informationsmodelle verstanden, deren Inhalte bei der Konstruktion anderer Informationsmodelle wiederzuverwenden sind [vom Brocke 2002; Alpar et al. 2002; Becker, Knackstedt 2004; Becker, Delfmann, Knackstedt 2004; Fettke, Loos 2004; Hafner, Winter 2008]. Die Wiederverwendung besteht in der Übernahme von Konstruktionsergebnissen sowie deren Anpassung und Erweiterung im anwendungsspezifischen Kontext. Die Intention ist es, durch die Wiederverwendung sowohl Effektivitäts- als auch Effizienzsteigerungen in der Modellierung, respektive der Informationssystementwicklung, zu erzielen. In der Referenzmodellierung werden Fragen der Konstruktion und Nutzung von Referenzmodellen thematisiert, um die Wiederverwendung der Modellinhalte möglichst wirtschaftlich zu gestalten.

Das zentrale Charakteristikum eines Referenzmodells ist seine intendierte bzw. faktische Wiederverwendung. Damit ist ein Referenzmodell ein Modell, das zur Wiederverwendung empfohlen oder faktisch zur Konstruktion weiterer Modelle wiederverwendet wird. Bisher hat sich innerhalb der Forschung zur Referenzmodellierung keine allgemein anerkannte Definition des Terminus Referenzmodell herausgebildet. Allerdings lässt sich in jüngster Zeit eine Konvergenz verschiedener terminologischer Bestimmungen in Richtung eines wiederverwendungsorientierten Referenzmodellbegriffs feststellen. Zur weiteren Systematisierung und unterschiedlichen Begriffsauffassungen siehe [Fettke, Loos 2004; Fettke 2006; Fettke, Loos 2007; vom Brocke 2003].

Business Process Models describe the activities of a company in an abstracted manner. Typically, the labeled nodes of a process model contain only sparse textual information. The presented approach uses an LSTM network to classify the labels contained in a business process model. We first apply a Word2Vec algorithm to the words contained in the labels. Afterwards, we feed the resulting data into our LSTM network. We train and evaluate our models on a corpus consisting of more than 24,000 labels of business process models. Using our trained classification model, we are able to distinguish different constructs of a process modeling language based on their label. Our experimental evaluation yields an accuracy of 95.71% on the proposed datasets.

Process model matching refers to the creation of correspondences between activities of process models. Applications of process model matching are manifold, reaching from model validation over harmonization of process variants to effective management of process model collections. Recently, this demand led to the development of different techniques for process model matching. Yet, these techniques are heuristics and, thus, their results are inherently uncertain and need to be evaluated on a common basis. Currently, however, the BPM community lacks established data sets and frameworks for evaluation. The Process Model Matching Contest 2013 aimed at addressing the need for effective evaluation by defining process model matching problems over published data sets. This paper summarizes the setup and the results of the contest. Besides a description of the contest matching problems, the paper comprises short descriptions of all matching techniques that have been submitted for participation. In addition, we present and discuss the evaluation results and outline directions for future work in this field of research

The matching of different process models and their nodes plays an important role, as shown in the manifold matching techniques developed during the last years. A well-established approach for the determination of those techniques’ quality is the consideration of precision and recall values related to a reference matching. Nevertheless, it is remarkable that it is not clear, what such a reference matching should be and how to reach a general acceptance. As developing a reference matching requires the decision and a consensus on what a matching should represent, we conceptualize that task as a decision problem, which can be well-structured or even ill-structured. The paper evaluates the evaluation of process matching techniques by theoretical argumentations and in terms of inquiring the process of reference matching development. Based on the results, the authors propose some guidelines containing the three phases idea, definition and criteria supporting that task.

Business Process Models describe the activities of a company in an abstracted manner. Typically, the labeled nodes of a process model contain only sparse textual information. The presented approach uses an LSTM network to classify the labels contained in a business process model. We first apply a Word2Vec algorithm to the words contained in the labels. Afterwards, we feed the resulting data into our LSTM network. We train and evaluate our models on a corpus consisting of more than 24,000 labels of business process models. Using our trained classification model, we are able to distinguish different constructs of a process modeling language based on their label. Our experimental evaluation yields an accuracy of 95.71% on the proposed datasets.

Reference models provide generic blueprints of process models that are common in a certain industry. When designing a reference model, stakeholders have to cope with the so-called ‘dilemma of reference modeling’, viz., balancing generality against market specificity. In principle, the more details a reference model contains, the fewer situations it applies to. To overcome this dilemma, the contribution at hand presents a novel approach to mining a reference model hierarchy from large instance-level data such as execution logs. It combines an execution-semantic technique for reference model development with a hierarchical-agglomerative cluster analysis and ideas from Process Mining. The result is a reference model hierarchy, where the lower a model is located, the smaller its scope, and the higher its level of detail. The approach is implemented as proof-of-concept and applied in an extensive case study, using the data from the 2015 BPI Challenge.

Reference modeling offers attractive bene®ts for both research and practice. The induc-tive strategy for reference model development derivesreference models by generalizing individual enterprise models. It has recently gained attention in research, however, its practical application still faces numerous challenges. The objective of the article at hand is to identify recent contributions to the ®eld of inductive reference model development and use them to analyze the current challenges that impede the application of their results in practice. We identify at otal of 18 contributions, either scienti®c articles describing inductive methods for reference model development, or practical reports describing the concrete development of areference model for acertain domain. Theyare all analyzed by means of asix-stage-framework for reference model development. Foreach stage, we derive speci®c challenges and point out acknowledgments and potential solutions. 1I ntroduction Reference modeling offers attractive bene®ts for both research and practice [FL07]. Following the epistemologically established differentiation between rationalism and empiri-cism as twofundamental ways of cognition, reference modeling differentiates adeductive and an inductive strategy for reference model development [BS97]. Model development according to the deductive strategy employs generally accepted theories and principles, while the inductive strategy is based on the generalization of individual enterprise models. It focuses on the commonalities of the individual models and abstracts from speci®c features. Hence, deductive development proceeds from the general towards the speci®c (ªtop-down "), whereas inductive development evolves from the speci®c into the general (ªbottom-up "). Although several concepts, methods, and tools for the support of reference modeling exist by now, practical development of reference models still faces avariety of challenges. The open questions range from project preparation, overindividual steps of pre-processing and derivation, to maintenance and continuous improvement of areference model. Hence, the development process of areference model today is often barely structured, nontransparent, and only marginally justi®ed and thus hardly communicable in terms of course and characteristics. In other words, the ideal of agenerally repeatable, engineering-style approach is usually not yet reached. With that, atool support is also only possible in alimited way.