Normalizing object-centric process logs by applying database principles

Much work has been done in process mining in the last two decades, where the focus of most efforts has been on unearthing the process models from log traces where each trace could be related to a unique case identifier that pertains to a single instance, such as an online customer order, a production order, a patient visit, etc. The case identifiers in these cases are customer order number, production order number, patient id, respectively, and there is a one-to-one relationship between the case identifier and the log data. On the other hand, in so-called object-centric (OC) logs, multiple objects are associated in one log record giving rise to many-to-many relationships among these objects and leading to ambiguities and redundancies in the log data. Hence, these logs become very difficult to analyze in their raw form as single linear files and it is important to convert them into database models. In this paper, we show how OC logs can be structured into a STAR and a fully normalized database schemas. The two schemas are compared and the benefits of our approach for log processing and ensuring log integrity are discussed.