Article Preview
Top1. Introduction
Nowadays, organizations need effective approaches to manage their business repositories that contain a large collection of business process models (Beheshti et al. 2016). For example, the repository of the Suncorp, which is an Australian bank and insurance company, contains more than 3000 models process models (Wang et al. 2014). An example of management activities conducted by these organizations are storing, merging, clustering, or querying Business Process models (i.e., BPs) (Beheshti et al. 2016).
Many of these management activities depend on a computed alignment by a matching process (Goncalo et al. 2015; Cayoglu et al. 2013; Jabeen et al. 2017). The process model matching process aims to find automatically correspondences between activities of two process models and its result is commonly referenced as an alignment. An alignment is a number of a pair-wise correspondences between two BPs activities sets. Each correspondence is a pair containing two similar sets of activities. The elements of the first set are activities of the first (i.e., source) business process model (i.e., BP), and the elements of the second set are activities of the second BP (i.e., target). (Euzenat & Shvaiko, 2007; Cayoglu et al. 2013).
Matching process models is widely recognized as a very hard task and many techniques have been proposed to solve this problem (Weidlich et al. 2010; Goncalo et al. 2015; Jabeen et al. 2017; Mostefai &Ignacio,2016; Mostefai & Mekkour,2020; Xue, 2019). The proposed approaches use a variety of techniques from different domains. For example, Weidlich et al(2013) used information retrieval techniques. In (Goncalo et al. 2015) the authors presented many approaches that combine various similarity metrics and Natural Language Processing techniques to solve the matching problem. Khurram et al (2019) evaluated the effectiveness of word embedding techniques such as Word2vec, GloVe and FasText, in solving the problem. The work of Xue (2019) studied the effectiveness of a compact memetic algorithm for process model matching. A machine learning approach to solve the problem is also presented by Sonntag et al. (2016).
Unfortunately, it was found that the proposed techniques are ineffective in aligning BPs. The literature shows also that no matcher can be of practical use since no one was able to outperform all the other matchers in all cases (Goncalo et al. 2015; Cayoglu et al. 2013; Jabeen et al. 2017). Thus designing effective tools to match effectively process models still an open problem(Jabeen et al. 2017).