The Study Of Detecting Difference Between Business Process Models

Nowadays,business processes are used by a lot of companies to manage their business operations,and these business processes are able to cross-border and cross-domain executed.The research on how to efficiently adjust the business processes to meet the complex market demands is important.Process difference detection is one of the most important techniques.Detecting differences in terms of structure,behaviore and resource is helpful for managers to analyze the performance of business operations.For example,two similar process models that belong to different companies are merged when two companies are consolidating in order to eliminate redundancy and enhance synergy.Thus,the managers need to identify their same and different parts,and determine which differences are applied to the merged process model.The challenges of process difference detection are listed as follows:Firstly,the existing process difference detection methods are incomplete and low efficiency in terms of node mapping.It is because they just map the task nodes while neglecting the non-task nodes.So how to completely and accurately map the nodes between two process models is a challenge.Secondly,the existing graph edit distance based methods are inefficient to compute the process difference,which are not suitable for detecting the complex process models.How to efficiently detect the differences between two process models is meaningful to companies.Thirdly,the edit script based difference is hard for users to understand,and it is difficult to extend the further analysis.How to visualize the differences and extend further analysis based on these differences,is an urgent problem to be solved.To solve the mentioned three challenges,on the basis of the existing process difference detection work,we first propose an improved node mapping method and then present three efficient process difference detection methods based on the node mapping.We conduct extensive experiments to show the practicality of the proposed methods in terms of effectiveness and efficiency.This paper contains the following three parts:Firstly,the existing node mapping methods just consider the task node and neglect the nontask node between two process models,which leads to a low accuracy of process difference detection.To solve this problem,we focus on the place node mapping based on the mapped task nodes in the Petri net based process models.The proposed place node mapping method is on the basis of the place context that consists of its input and output task nodes,which greatly improve the mapping accuracy.The experiment results show that the performance of the proposed method can meet the requirements of the practical application in terms of effectiveness and efficiency.Secondly,since the existing graph edit distance based process difference detection methods are not suitable for the complex process models,we propose two efficient process difference detection methods.The first method computes a dissimilarity value between two process models to measure their different degree.The dissimilarity value is equal to 1 minus their similarity,the larger this value is,the more different these two process models are.First,we design the place similarity measurement based on the place context.Then,the similarities of all possible combinations of places are computed based on the place similarity measurement,and the Hungarian algorithm is used to determine the optimal place combination among them.Next,the process similarity is calculated based on the optimal place combination.Finally,we obtain the dissimilarity value on the basis of this process similarity.Experiment results show that our proposed method achieves the same precision as the baseline methods.Besides,our proposed method outperforms the baseline methods in terms of efficiency.The second method is to transform two process models to task-based process structure trees(TPSTs),and compute their approximate minimum edit script,where this edit script is regarded as the approximate optimal difference between two process models.We first split the TPSTs into fragments,where each fragment is represented by a feature vector.Then,the mapped fragments in two TPSTs are obtained by calculating the similarities between their corresponding feature vectors,and the node mapping is determined in each mapped pair of fragment.Finally,the approximate minimum edit script is calculated based on the fragment mapping and the node mapping.Experiments show that our proposed method can be used in real application scenarios.Thirdly,since the edit script based difference is hard to understand and extend the further analysis,we first summarize a set of process difference patterns.Then a difference detection method is designed to visualize the process difference in different levels of abstraction based on the difference patterns.The difference patterns can explicitly display the structural and behavioral differences,and the cost and time based process difference analyses are also easy to extend based the difference patterns.Based on the multi-level process difference visualization,users can not only globally understand the process difference,but also focus on the difference in a specific level of abstraction.A case study from real life and extensive experiments show that the proposed method can be applied to the practical application.