Extracting workflow structures through Bayesian learning and provenance data

Mining workflow models has been a problem of interest for the past few years. Event logs have been the main source of data for the mining process. Previous workflow mining approaches mostly focused on mining control flows that were based on data mining methods, as well as exploited time constraints of events to discover the workflow models. In this work, we present a mining approach which not only takes the behaviourial aspect of workflows into account, but also takes advantage of their informational perspective. Provenance information is a source of reasoning, learning, and analysis since it provides information regarding the service inputs, outputs and quality of service values. Therefore, provenance information along with Bayesian structure-learning methods are exploited for this purpose. Two constraint-based Bayesian structure-learning algorithms are investigated and modified in order to make use of additional provenance information. We will show that this leads to better mining results based on three common mining scenarios.