| With the increasing market competition,more and more enterprises use the behaviors recorded in the information system to establish business process,and they hope to achieve better goal.Some behaviors in the initial business process will change or generate new behaviors due to the constant change of the balance between the actual needs and interests in the development process of enterprises.It is important for the management and optimization of business processes that enterprises can take the initial behavior and changing behavior of business processes in actual development into account.In process mining,the current behavior recorded in the information system is regarded as an event log,and the initial behavior of the information system can be constructed as a model by process discovery.The differences between event log and initial process model are discovered by associating them.Therefore,detection and repair deviation between event log and initial process model has always been an important topic in the field of process mining.This topic is studied on the basis of the existing methods of deviation detection and model repair.The main contributions can be summarized as follows:(1)Aiming at the problem of deviation detection between event log and process model,the event log and process model are directly aligned or the optimal alignment is only determined based on the number of deviation occurrence by existing method.So the alignment optimization cannot be analyzed reasonably according to the actual situation.In this paper,the optimal alignment is determined by eliminating algorithm and the total cost of the deviations,and the size of optimal alignment or the possible impact on the precision are taken as additional measurement criteria.Different cases were used to evaluate the performance.The following two results could be obtained when fitness was prior in alignment processing:<1>Compared with direct alignment,the fitness could be improved in unweighted case sets generated by BPIC2020 and real business processes,respectively;<2>Compared with the A*algorithm,the fitness of seven weighted cases and an unweighted case could be improved.Compared to the A*algorithm,the precision replay value could be improved in an unweighted case when fitness was prior in alignment processing.(2)Aiming at the problem of repairing the deviation between event log and process model,existing method adopted the self-loop to insert each deviation element generated in event log into the initial process model,independently.Although the fitness after repair guaranteed to be 1,the precision was sacrificed as a cost.In this paper,the consistency check of behavior relationship is combined with the deviation detection of optimal alignment.The occurrence,location and potential behavior relationship of deviation are detected in optimal alignment based on the unfitted behavior pattern.Then,the deviation elements with directly following relationship in event log are constructed into a substructure satisfied with specific behavior relationship.The Mrepair-Check*plug-in was used to verify the repair performance.The result showed that the size of the event log could be reduced by constructing the deviation substructure,so as to the precision could be effectively improved under the premise of fitness 1.(3)Aiming at the problem that two different deviations in event log have different replay forms,existing methods adopted a unified way to repair all deviations in the log,which often is hard to reconcile both fitness and precision.The current variant between the event log and the process model is found in real time in this paper.Then,it need to be checked whether the iterative deviation generated by event log can exist or not in this variant.The repair or configuration operation is selected according to the check result.The experiment was evaluated by the Mrepair-Check plug-in.The result of the overall repair performance showed that more insert deviations could be accurately replayed by configuration operation of specific variant.Thus the precision was improved as much as possible under the premise of ensuring the fitness.The result of real-time repair performance showed that the fitness and precision of the proposed method were always 1 in the variant without iterative insert deviation,but the existing method required a large amount of precision to ensure the fitness.(4)Aiming at the problem of reasonable trade-off between fitness and precision of model repair,existing method set the ideal fitness after repair to a fixed value of 1,so excessive pursuit of fitness may lead to unreasonable precision in some cases.The whole optimized fitness using configuration operation is predicted based on the total cost of iterative deviations in the event log in this paper.In the case that the predictable fitness meets a reasonable range,configuration operation is used to repair all deviations.Otherwise,the repair or configuration operation is performed based on the check result of the iterative observable deviation in the current variant.The experiment was evaluated by Mrepair-Check*and Mrepair-Check plug-ins.The result showed that the real-time repair based on alternative operations could make the final fitness as 1 when the predictable fitness was reasonable,and the precision of one data set was 1%lower than the reasonable range.However,the proposed method can ensure that the fitness and precision are always reasonable.The performance of this proposed method is consistent with that of the real-time repair based on alternative operations when the predictive fitness was not reasonable.Their precisions are 8%higher than that of the repair based on deviation substructure in the six data sets without loop paths.In conclusion,the deviation information of various business processes under different circumstances can be reasonably detected by the proposed method.At the same time,the model repair method improves the repair performance through layer by layer,and the precision is effectively improved under the fitness is 1 or within a reasonable range.Figure[71]:Table[50]:References[120]... |
PDF Full Text Request