An Extended α-Algorithm To Discovery Complex Loops In Process Mining

Process mining can analyze event logs generated by enterprise applications to reproduce the actual execution of business processes, and obtain objective and valuable knowledge to conduct optimization and reconstruction of workflow models. It's a new application of data mining in workflow management. Since 1998, Agrawal and other researchers who work in IBM Almaden Research Center firstly proposed a technique which can construct workflow models from event logs, there has been a great deal of research on process mining. And the identification and solution of complex loops in workflow models are still a key and difficult issue in process mining. Therefore, the study of process mining algorithms, which have the ability of mining complex loops from event logs, is a new direction and challenge in future work of process mining.Based on a-algorithm and its series of extended algorithms, this paper analyzes the algorithm have the disadvantage of not mining complex loops from event logs. To address this issue, the paper discusses and analyzes the workflow traces which generated by the different types of nested structures of loops and other model structures, proposes a recognition method according to heuristic decision rules based on the causal dependencies between activities to discover and deal with complex loops in workflow models.Utilizing a recursive definition which includes causal dependencies between activities and traces of loops, the recognition method is applied in a-algorithm to improve its ability to mine complex loops. The extended a-algorithm works as follows:First, identify records produced by the outermost loops and replace them temporarily with virtual tasks to extract traces of loops from event logs, and then apply a-algorithm to mine main workflow model from the event log without traces generated by the outermost loops; Second, mine sub-models of the outermost loops from the interrelated traces and replace the corresponding virtual task of the main workflow model for each sub-model; Finally, conduct the previous process recursively to mine sub-models from traces generated by loops and add all sub-models to the main workflow model until get the complete model.The extended a-algorithm is implemented and integrated locally as a mining plug-ins of the open-source framework for process mining named ProM. In addition, through theoretical analysis and simulation experiments, the paper conducts a comprehensive experimental evaluation of the extended a-algorithm and the conclusion can be drawn as that the capacity of mining complex loops in workflow model of extendedα-algorithm is feasible and more effective than that ofα-algorithm.