Approximate Clone Detection in Repositories of Business Process Models

Abstract

Abstract. Evidence exists that repositories of business process models used in industrial practice contain significant amounts of duplication. This duplication may stem from the fact that the repository describes variants of the same processes and/or because of copy/pasting activity throughout the lifetime of the repository. Previous work has put forward techniques for identifying duplicate fragments (clones) that can be refactored into shared subprocesses. However, these techniques are limited to finding exact clones. This paper analyzes the problem of approximate clone detection and puts forward two techniques for detecting clusters of approximate clones. Experiments show that the proposed techniques are able to accurately retrieve clusters of approximate clones that originate from copy/pasting followed by independent modifications to the copied fragments.

A Short Survey on Process Model Similarity

“…This will provide a good basis for further comparative studies such as [33]. Related application scenarios such as clustering process models [34], detection of clones [35] and behavioural patterns [36] are expected to further benefit from research into process model similarity.…”

Controlled automated discovery of collections of business process models

“…In addition to allowing us to identify exact clones, the RPSDAG provides a basis for approximate clone detection [20]. Approximate clone detection is achieved by applying clustering techniques on the collection of SESE fragments of an RPSDAG, using one minus the graph-edit distance as the similarity measure (as defined in [21]).…”

“…Two clustering techniques for approximate clone detection based on this principle are presented in [20]. The first is an adaptation of the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, the second is an adaptation of the Hierarchical Agglomerative Clustering (HAC) algorithm.…”

“…For each set of exact clones, we create a single subprocess and replace the occurrence of these clones in their process models with a subprocess activity pointing to the subprocess just created (lines 6-7). Once exact clones have been factored out, we identify clusters of approximate clones using the technique in [20] (line 8). For each fragment cluster, we merge all approximate clones in that cluster into a configurable fragment (line 11) using the technique in [22].…”

“…Thus, we have to order the processing of the approximate clones clusters based on some benefit-cost-ratio (BCR), so as to prioritize those clusters that maximize the number of process models satisfying the threshold after approximate clones extraction (line 10). If we set our threshold on size, we can use the BCR defined in [20], which is the ratio between overall size reduction (benefit) and distance between approximate clones within a cluster (cost). Similar BCRs can be defined on other complexity metrics.…”