The Research And Implementation Of Workflow Process Mining Algorithm Based On Event Multiset

Nowadays, most of information systems use the well-defined workflow models to describe task relations and manage whole business process. However, with the increasing of business processes and the process become more complicated, the workflow models have the inefficient, even the defect problem. Therefore, it is necessary to monitor and improve the business process which needs to obtain the reality process of workflow models. As the automation systems of centralized management workflow, workflow systems are usually composed by a variety of workflow. The workflow is based on a series of procedures or rules of the documents, information or activity which transferred from one participant to another in the whole or part of the business process. The goal of workflow mining is to rediscovey runtime business process model from the workflow log generated by the workflow system, and support the design and analysis of workflow technology.In the prodecure of research, we find that the classic workflow mining algorithms are based on the event trace, i.e. event trace is the unit of analysis, this kind of algorithm can be called mining algorithm based on event trace. Typical steps of this kind of algorithm can be described as:(1) generating event traces from event log;(2) analyzing event traces and obtaining ordering relations of tasks;(3) generating process model with ordering relations of tasks. Because the step (1) and (2) needs to analyze all of events in the trace, these two phases is a major performance consumption of this algorithm. However, the workflow mining algorithm proposed in this paper, i.e., λ-algorithm, is not based on event trace, but event multiset which is called the mining algorithm based on event multiset.λ-algorithm doesn’t need to analyze the event trace which is able to improve the efficiency of mining algorithm. In addition, the log scale growth of event multiset trends to a constant value, but the log scale growth based on event trace is linear. Hence, the former log scale is much smaller than the latter which further enhance the efficiency of A-algorithm. λ-algorithm is able to obtain the process model more efficiency.As the source data of λ-algorithm, event multiset is composed by a variety of reusable event set. On the one hand, each event in event multiple is composed by current task and post-task, i.e., with post-task of event; On the other hand, event multiset doesn’t include workflow instance number which isn’t care of event trace,λ-algorithm is able to directly discovery causal dependency and parallelism dependency by analyzing the relationship between current task and post-tasks,λ-algorithm is not only able to correctly mining the SWF-net with short loop structure, but also is able to mining implicit dependency structure, implicit dependency and some non-well-handled nets. λ-algorithm is able to handle a wider range of workflow nets.The author has developed the plugin ofλ-algorithm "Lamda Miner Process Multiset Plugin" which is based on the ProM framework (the open framework of workflow). The log format of this is the mul (event multiset log). Mining a complex software project management process with this plugin, the mine result is correct and the performance is efficient. This paper also makes the detail experimental evaluation with λ-algorithm plugin in the aspect of conformance and performance.