Applying Embedding Methods to Process Mining

Abstract

The performance of process mining algorithms on a particular event log highly depends on the number of different processes present in the logs. Prior event log clustering can help find out which processes certain events in the logs belong to. Since log clustering is not always a simple task, a preliminary transition from logs to log embeddings can be an important step in solving process mining problems. In this paper, we apply different embedding methods to a dataset of event logs. By transitioning to log embeddings and applying clustering methods we improve the efficiency of process mining. The experiment results suggest that embeddings capturing events order perform better than others.