Prof. Dr. Hendrik Leopold ©Hendrik Leopold

Forschungskolloquium Wirtschaftsinformatik

06. Mai

Im Rah­men des For­schungs­kol­lo­qui­ums Wirt­schafts­in­for­ma­tik und Data Sci­ence re­fe­riert am 06.Mai.2021 um 16:15 Uhr Herr Prof. Dr. Henrik Leopold, ordentlicher Professor an der Kühne Logistics University (KLU) und außerplanmäßiger Professor am Hasso-Plattner-Institut (HPI) an der Fakultät für Digital Engineering der Universität Potsdam, virtuell über "Towards Conformance Checking without Process Models: Detecting Execution Anomalies in Event Logs Using Natural Language Processing".

In many domains it is important that the execution of business processes adheres to certain rules and regulations. To efficiently detect violations against such rules and regulations, conformance checking techniques have been introduced. They compare the actual behavior of employees, as recorded by information systems, to the desired behavior that is specified in a normative process model. In this way, non-conforming behavior can be automatically detected and respective measures for their prevention can be introduced. While these techniques have been found to be valuable in many settings, they are only applicable if a process model capturing the normative process behavior is available. Recognizing that this is often not the case in practice, several authors have developed so-called anomaly detection techniques. A key assumption of all these techniques, however, is that anomalous behavior is less frequent than desired behavior. In this talk, we argue that such a frequency-based perspective is too limited since it ignores the semantics of the recorded events and, therefore, does not take the meaning of potentially anomalous patterns into account. To bridge this gap, we put forward the idea of exploiting a, so far, disregarded dimension of event logs for anomaly detection: the natural language from event labels. We build on the linguistic branch of semantics, which is concerned with the meaning of words. Our key idea is that anomalous process behavior can be identified based on the detection of semantically inconsistent execution patterns. As an example, consider a process instance in which an order is both accepted and rejected. From a semantic point of view, this is an undesirable constellation since accepted and rejected are opposites (so-called antonyms).

Zugangsdaten

https://leuphana.zoom.us/j/96455201912?pwd=bzBwT3NqL3o2b3RqRGZwL2VwS0RBdz09

Meeting-ID: 964 5520 1912
Kenncode: 5ghwu8