Process Model Matching with Word Embeddings

Abstract

Business process models are widely pronounced as valuable assets for every organization. Given that manual management of these process models requires substantial human effort, process model repositories have been developed to effectively manage these models. The key features of these repositories include process model searching, clone detection, duplicate avoidance, and process harmonization. The usefulness of all these features depends upon the accuracy of the underlying process model matching technique. Process Model Matching (PMM) refers to identifying corresponding activities between a pair of process models that represent the same or similar functionality. Recognizing the importance of PMM, a plethora of matching techniques have been developed. Despite the presence of these techniques, the need for enhancing the accuracy of PMM have been widely pronounced in recent studies. To that end, this paper proposes a word embeddings based approach to enhance the accuracy of PMM. For the evaluation of the proposed approach, we have used three state-of-the-art word embeddings, Word2vec, Glove, and fastText, for experiments on three benchmark datasets. The results show that the fastText generated embeddings, that are recently released by Facebook Inc., are the most suitable embeddings for process model matching.