Research On Business Process Similarity Based On Event Log

The business process is the basis of enterprise operation and an important driving force for the normal operation of enterprise business.With the continuous emergence of various new management concepts,new technologies and new methods,enterprise management has gradually shifted from functional management to process management.This not only brings various new challenges to process management,but also puts forward a series of more complex,diverse and higher requirements for the measurement and calculation of business process similarity.At present,the research on business process similarity mainly focuses on two aspects:Firstly,how to use the behavior information in the log has become the key problem to improve the accuracy of business process similarity measurement;Secondly,as an important input of similarity measurement,how to ensure the completeness of logs is also of great significance.In order to solve the above two problems,from the perspective of the model and the log,this paper combines the behavioral relations in the model with the behavioral relations in the log to improve the accuracy of similarity calculation between business processes.At the same time,the trace clustering method is used to repair the log to ensure its completeness.Based on the behavioral relations between activities,this paper makes the similarity measurement between processes more accurate and closer to reality through the two-dimensional differentiation of variants with equivalent distance and the combination of the model and the log.The main research is as follows:(1)A discrimination method of variants with equivalent distance based on quasi behavioral profile is proposed.On the basis of equal edit distance,the similarity of quasi behavioral profile is introduced,and the edit distance is combined with the similarity of quasi behavioral profile to realize the two-dimensional distinction between business processes.Finally,six variants with equal distance are further distinguished by using quasi behavioral profile similarity,which solves the problem that a single dimension cannot distinguish variants.(2)A similarity calculation method based on model behavior and log behavior is proposed.Firstly,based on the similarity matrix between behavioral relations,the similarity of model-based behavior is defined.Then,the concepts of occurrence identification and significant frequency are given,and then the similarity based on log behavior is defined though reasonably quantifying the frequency of the behavioral relation in log by statistical methods.Combining the two similarities,MLB similarity formula is proposed to calculate the similarity between the two processes.The experimental results show that this method can correctly calculate the similarity between business processes.(3)An event log repair method based on model clustering is proposed.Firstly,all complete traces are clustered with the subgraph of the model as the centroid;Secondly,the missing trace is matched with the candidate cluster according to the alignment cost to each centroid and the distance to other traces in the cluster;Then,the missing trace is repaired according to the repair algorithm to obtain the complete trace.Finally,the feasibility and effectiveness of the proposed method are verified by experimental simulation.Figure [14] Table [15] Reference [97]...