Efficient Retrieval And Parallelism Refactoring Of Process Models

With the wide use of business process management technology, there are more andmore process models in diferent industries. Business process models describe how busi-ness is processed, and they can be used to drive process aware information systems. Sincebusiness process models are very important assets for organizations, how to manage sucha large number of process models is challenging. We look on these process models asa kind of unstructured data, and have done some research work on retrieval and qualitymaintenance as follows.Model retrieval: With the help of model retrieval, process modelers need not todesign a new model from scratch, they can find some related models and then workon them. It would save a lot of time and is less error-prone. When some smallcompanies are merged into a bigger one, we need to integrate similar business, andsimilar process model retrieval can facilitate this work. Since the behavior spaceof a process model is large in most cases, and the number of process models isbig in large enterprises, the efciency of retrieval is important. We propose newefcient algorithms to compute the behavior of process models based on unfoldingtechnology. Since process model have both structure information and behaviorsemantics, and the retrieved models can satisfy the users’ requirement exactly orsimilarly, we classify the model retrieval problem into four diferent categories.To facilitate the efciency of model retrieval, we propose four diferent indexesto support diferent types of model retrieval. Since diferent modelers would usediferent labels for the same tasks, we consider label similarity between tasks, andwith the help of label index, the efciency of model retrieval is still high.Refined task relations computing: The behavior semantics of process modelsdescribe how business is processed. When we retrieve process models based onbehavior, we need to describe the execution relations between tasks more preciselyso that the resulting models can satisfy the users’ requirements more precisely.When we check whether a given model comply with the given business rules, wealso need to describe the execution relations between tasks more precisely. Basedon whether the two given tasks are always executed in the same instance, we refinethe causal relations and parallel relations, and propose some theorems to compute the refined relations quickly. All the theorems are proved. For acyclic processmodels, we propose an algorithm to calculate the refined relations between tasks.Parallelism refactoring: Since process models describe how business is pro-cessed, process models themselves determine the efciency of business processingto a great extent. If the tasks in a process model can be executed in parallel asmuch as possible, the efciency of the corresponding business processing wouldbe raised much high. For the end of parallelism refactoring, we analyze the dataprocessing during business processing, change some tasks in sequence structuresinto parallel structures. With the help of process mining technology, we refactortasks into parallel structures as much as possible, so that business can be processedas much efciently as possible. In other words, the quality of process models canbe improved with parallelism refactoring.We have implemented above approaches in BeehiveZ, an open source system, and exper-iments show that our approaches are efective and efcient.