Business Process Modeling Abstraction Based on Semi-Supervised Clustering Analysis

The most prominent Business Process Model Abstraction (BPMA) use case is the construction of the process “quick view” for rapidly comprehending a complex process. Some researchers propose process abstraction methods to aggregate the activities on the basis of their semantic similarity. One important clustering technique used in these methods is traditional k-means cluster analysis which so far is an unsupervised process without any priori information, and most of the techniques aggregate the activities only according to business semantics without considering the requirement of an order-preserving model transformation. The paper proposes a BPMA method based on semi-supervised clustering which chooses the initial clusters based on the refined process structure tree and designs constraints by combining the control flow consistency of the process and the semantic similarity of the activities to guide the clustering process. To be more precise, the constraint function is discovered by mining from a process model collection enriched with subprocess relations. The proposed method is validated by applying it to a process model repository in use. In an experimental validation, the proposed method is compared to the traditional k-means clustering (parameterized with randomly chosen initial clusters and an only semantics-based distance measure), showing that the approach closely approximates the decisions of the involved modelers to cluster activities. As such, the paper contributes to the development of modeling support for effective process model abstraction, facilitating the use of business process models in practice.