Master's thesis, University of Innsbruck, Institute of Computer Science, 2016.
Business process modeling became increasingly popular over the last years. Using systematic techniques, organizations document their business processes. In this respect, it is important to ensure a certain degree of quality of process models, which is, however, problematic to ensure. The quality of process models can be determined on a syntactical level as well as on a semantical level. So far, previous research mostly investigated the syntactical quality of process models (e.g., soundness property of process models).
The semantic quality of process models is influenced, among others, by the quality of process labels (i.e., the label or name of a graphical construct) as these process labels convey the meaning of the graphical constructs and thus the meaning of the process model. The quality of process labels is determined, for instance, by the label style, label length, and the degree of detail provided by the label.
This thesis presents a system that focuses on the quality of process labels. Following the design science research paradigm, a system was created that is, first, able to automatically detect the label styles of process labels. Second, based on process labels, the system identifies modeling weaknesses and provides recommendations on how to best address the identified weaknesses. Overall, the system provides recommendations for seven modeling errors.
In two evaluations, both functions of the system were tested. Using pre-classified process labels, the functionality for detecting label styles was evaluated. In this evaluation, the system reached sensitivity and specificity values of up to 100%. On average, the sensitivity values reached 91% and 84% and specificity reached 98% and 97% for German and English labels, respectively. The recommendations were evaluated with experienced business process modelers. This evaluation showed that they perceive the recommendations helpful to remedy modeling errors. For German process labels the recommendations were perceived as helpful in 83% cases, for English labels this value equals to 81%. Overall, the recommendations were perceived as helpful in 82% cases.
More publications by