Извлечение читаемых моделей из логов событийDiscovering readable models from event logs

Modern information systems record detailed process execution histories in event logsto monitor their behavior and improve processes. These logs allow process miningmethods, but the resulting models are often too complex and difficult for humanexperts to understand due to the large amount of input data. The dissertationaddresses the challenge of creating readable hierarchical process models from lowlevelevent logs while ensuring their conformance with observed behavior.The key contributions of this work include a conformance checking algorithmfor high-level process models and low-level events logs. The algorithm has beenproven to be correct for perfect fit and is robust against noise. Another contributionis a hierarchical model discovery algorithm that synthesizes abstract models fromdetailed logs, supporting cyclic and concurrent behavior, enabling the creation ofstructured hierarchical workflow nets. Experimental validation on synthetic andreal logs demonstrates that hierarchical models are more compact and readable thanclassical models for large systems, while maintaining high fitness and precision metricvalues. The proposed methods bridge the gap between the system-level event log anda more readable process model for experts, facilitating better analysis of businessprocesses.