Simplifying process model abstraction: Techniques for generating model names

Highlights

• We proposes a technique to automatically infer suitable names for business process models and fragments thereof.

• This technique is useful for model abstraction scenarios and refactoring.

• The technique is grounded in an adaptation of the theory of meaning.

• We implemented the technique in a prototype tool.

• We conducted an extensive evaluation using three process model collections and a case study.

Abstract

The increased adoption of business process management approaches, tools, and practices has led organizations to accumulate large collections of business process models. These collections can easily include from a hundred to a thousand models, especially in the context of multinational corporations or as a result of organizational mergers and acquisitions. A concrete problem is thus how to maintain these large repositories in such a way that their complexity does not hamper their practical usefulness as a means to describe and communicate business operations. This paper proposes a technique to automatically infer suitable names for business process models and fragments thereof. This technique is useful for model abstraction scenarios, as for instance when user-specific views of a repository are required, or as part of a refactoring initiative aimed to simplify the repository's complexity. The technique is grounded in an adaptation of the theory of meaning to the realm of business process models. We implemented the technique in a prototype tool and conducted an extensive evaluation using three process model collections from practice and a case study involving process modelers with different experience.

Introduction

Recent years have seen a substantial increase in business process modeling initiatives. While process modeling was utilized in the 1990s mainly as a technique for facilitating single process re-engineering efforts [1], [2], [3], [4], [5], many companies have turned to a more encompassing and evolutionary approach to Business Process Management (BPM). This development has led to the establishment of BPM expert teams, competence centers, and consulting departments within organizations. These units typically manage a central repository of business process models capturing various aspects of an organization's business operations. Process modeling is an ongoing activity in a setting that is interwoven with strategic management, quality assurance, controlling, and the specific functional areas of a company. As a result, companies that model business processes often maintain repositories containing hundreds if not thousands of models at a remarkable level of detail [6].

The mass of documentation stored in a process model repository poses considerable challenges for efficient and effective use of these models. At various stages, different stakeholders require individual views on these models for getting an overview, receiving support during their modeling efforts, and for quality assurance purposes. For example, abstract views are a good vehicle for achieving a cognitive fit between the representation of the process and the task at hand, cf. [7], [8]. Moreover, they offer a suitable aggregation of process information, which has been found to be crucial for process improvement initiatives [9].

A survey on how process models are used in practice reveals no less than fifteen different use cases for abstracting and compressing a detailed, fine-granular model toward a smaller model capturing the essential information a stakeholder is after [10]. These use cases clearly motivate the development of techniques that facilitate such transformations, e.g., [11], [12], [13]. These techniques, however, do not address the problem of how to automatically generate a proper name for an abstract model. This puts the burden on the stakeholder to come up with a meaningful reference for each new abstract model that is derived. This is a task that cannot be ignored, since much of the meaning of a process model can only be derived from what its elements stand for. It is also potentially a highly repetitive task: abstractions for a single process model may be dynamically generated requiring a name proposal each time they are viewed by a stakeholder. The same problem occurs when searching for similar or identical model fragments in large process model repositories. Such “(approximate) clones” can be automatically found [14], [15], [16], [17], but then need to be stored with a meaningful name. An automatic technique for determining the name of a process model clone is not yet available, thus hindering the wider adoption of clone detection in practice. In this paper, we address the problem of defining names for process models and process model fragments that, for instance, result from applying abstraction or clone identification. Our contribution is an approach that will ease the naming task for users who are interested in creating more abstract views on process models than are readily available to them. Specifically, our approach builds on techniques that analyze the activity labels of a process model, and returns a ranked list of names from which the top ranked name is provided. To this end, we build on insights from theories of meaning and exploratory research to devise naming strategies for process models. As we will argue, the quality of the generated naming suggestions is comparable to those that would be generated by humans; however, in the proposed approach these suggestions are generated in only a fraction of the time a human modeler would need to inspect the underlying (fine-grained) model elements and their interrelations.

We implemented the naming approach in a prototype tool based on natural language processing techniques, and, in order to demonstrate the flexibility of our approach, we configured the tool to detect names for process models specified in two languages: English and German. Next, we validated the approach in various directions. First, we measured the performance and accuracy of our approach using three large datasets from practice. Second, we conducted a case study involving process modelers with varying expertise to evaluate the usefulness and appropriateness of the approach in practice.

The rest of the paper is structured as follows. Section 2 reviews the essential concepts of business process modeling, presents a taxonomy of naming strategies grounded in theories of meaning, and highlights an exploratory study on process model names. Section 3 introduces the approach for automatically deriving process model names. Section 4 discusses the results of the evaluation while Section 5 summarizes achievements, and suggests implications for research and practice. Section 6 concludes the paper with an outlook on future research.