The goal of process mining is to obtain objective and valuable information fromevent logs. The research of process mining is of great significance for deploying newbusiness processes as well as analyzing and improving the already deployed ones. Pro-cess mining consists of mining various perspectives of a process model, such as thecontrol ?ow perspective, the organization perspective, the case perspective, etc. Thethesis is focused on mining the control ?ow perspective of a business process and givessolutions for the two open problems of mining process models with non-free-choiceconstructs and invisible tasks. The key contributions of the thesis are as follows:1. Towards such event logs including both start and complete events of a task, theβalgorithm, whose correctness can be proved theoretically, is proposed to minesound structured work?ow nets from them.2. The indirect dependencies between tasks in process models are classified intodifferent categories for the first time. Then the inter-task binary relationships thatcan be found from event logs are extended accordingly. Based on the differentstructural characteristics of the indirect dependencies, several detection methodsare introduced, whose correctness can be theoretically proved. The proposedα++ algorithm is able to successfully mine sound work?ow nets containing non-free choice constructs from event logs.3. Based on the functional classifications of the ordinary invisible tasks, theα#algorithm is proposed, which is able to mine all the four kinds of invisible tasks.Theα# algorithm makes use of several reasoning methods for the detection ofinvisible tasks, whose correctness can be proved theoretically.The algorithms mentioned above are implemented as mining plug-ins of theframework for process mining named ProM. A lot of artificial logs and real-life logsfrom corporations are used to evaluate the effectiveness of the proposed algorithms. |