Parameterised trace selection technique for process model recovering

Business processes are often implemented in software applications which expose them by means of a Web interface. Several process recovery techniques apply dynamic analysis for tracing application runs and collecting information used to infer a model of the implemented application process. However, the recovered process model can be very complex, intricate and difficult to understand. Indeed, as often happens when considering dynamic analysis for model inference, recovery techniques try to describe all possible application executions. Some of the executions, however, can vary only for minor changes, thus resulting not particularly relevant for documenting the underlying application process. In this study, the authors propose the use of trace selection for improving the understandability of the recovered process model. The authors investigate the use of a trace-clustering technique to reduce the considered set of traces, selecting those traces that are representative of the whole set, while discarding those capturing minor variants of main behaviours.