Detecting Approximate Clones in Process Model Repositories with Apromore

Approximate clone detection is the process of identifying similar process fragments in business process model collections. The tool presented in this paper can efficiently cluster approximate clones in large process model repositories. Once a repository is clustered, users can filter and browse the clusters using different filtering parameters. Our tool can also visualize clusters in the 2D space, allowing a better understanding of clusters and their member fragments. This demonstration will be useful for researchers and practitioners working on large process model repositories, where process standardization is a critical task for increasing the consistency and reducing the complexity of the repository. 1 Overview of the tool Identification and analysis of similar process fragments, aka approximate clones, is a major step in business process standardization initiatives, where similar process fragments can be replaced with standardized fragments to reduce differences across different organizational units, products or brands. In order to offer concrete support to such process standardization initiatives, we developed a tool that allows analysts to identify, cluster, analyze and visualize approximate clones. The tool is part of the Apromore advanced process model repository [5, 3]. The purpose of Apromore goes beyond that of simple model storage. Apromore aims to provide a one-stop place for the research community to expose algorithms and techniques that operate over (large) process model collections. Examples of techniques that have already been implemented are process similarity search [1] and process merging [4]. An advantage of being integrated into Apromore, is that the tool exploits Apromore’s canonical process format, an independent format used for internal process representation. All process models imported into Apromore are converted into this internal format. Doing so, approximate clones can be detected in process models defined in different modeling languages such as BPMN, EPC, PNML, etc. Apromore is a SaaS reachable via the Web. The functions offered by the approximate clone detection tool are available through Apromore’s Web interface (the Apromore portal), as well as via Web service operations. The Apromore portal consumes ? Work done while visiting Queensland University of Technology, Australia 2 C.C. Ekanayake et al. these operations itself, but they can also be consumed by external applications (e.g. the WoPeD tool5 – a Petri net editor – can connect to Apromore). Fig. 1. Web interface of the approximate clone detection tool in Apromore The Web interface of the approximate clone detection tool (shown in Fig. 1) provides features for creating, browsing and visualizing fragment clusters. Users can select one or more process models, specify the clustering parameters (such as the preferred clustering algorithm), and kick off the clustering. Once the fragments included in the selected process models have been clustered, users can apply different filtering criteria (e.g. on the size of the clusters) and browse the resulting clusters in a detailed list view. Another useful feature is the visualization of clusters in the 2D space. The visualization component (shown in Fig. 2) displays each fragment in a cluster as a point in the space and positions fragments within a cluster according to their distances to the medoid (distances being represented as edges between the points). It also positions the clusters in the space according to the GEDs among their medoids. One can also click on the point corresponding to a process fragment and visualize its corresponding model using any process modeling language supported by Apromore (e.g. EPCs, BPMN). Under the hoods, the approximate clone detection tool relies on three techniques that have also been integrated into Apromore: i) RPST, ii) RPSDAG and iii) graph-edit distance. The RPST algorithm [6] is used to decompose each process model into a set of Single-Entry Single-Exit (SESE) process fragments. Such decomposed process fragments and their parent-child relationships are stored in the RPSDAG [7], an indexing structure which captures the union of the RPSTs of all process models by identifying cloned process fragments. This information about fragments and their parent-child relationships is used by a clustering algorithm to identify meaningful clusters.