Inferring Missing Entity Identifiers from Event Knowledge Graphs

Complete event data is essential to perform rich analysis.
However, real-life systems might fail in recording the (correct) case identifiers
the system has operated on, resulting in incomplete event data. We
aim to infer missing case identifiers of events by addressing the physical
constraints of its process which previous work has failed to do. We modelled
event data and its physical context in an Event Knowledge Graph
(EKG) and formalized a definition for inference rules using EKGs. Five
inference rules regarding physical objects are created to infer identifiers
in a synthetic data set and a data set from the IC manufacturing industry.
The approach is evaluated using conformance checking. Initially,
none of the traces were complete. Using our method, we could infer a
case identifier for 95% of the events resulting in 88% complete traces.