Mining for Long-Term Dependencies in Causal Graphs

Process discovery is one of the most challenging tasks in process mining. Based on event data, a process discovery approach generates a process model that captures the behavior recorded in the data. The hybrid miner is a two-step process discovery approach that creates a balance between the advantages of formal modeling and the necessity of remaining informal for vague structures. In the first discovery step, an informal causal graph is constructed based on direct succession dependencies between activities. In the second discovery step, the hybrid miner tries to convert the discovered dependencies into formal constraints. For vague structures where formal constraints cannot be justified, dependencies are depicted informally. In this paper, we reduce the representational bias of the hybrid miner by exploiting causal graph metrics to mine for long-term dependencies. Our evaluation shows that the proposed approach leads to the discovery of more precise models.