Improving Process Discovery Using Translucent Activity Relationships

In task mining, data captured by recording user interactions in a desktop environment are stored in a user interaction log. A user interaction log can be transformed into an event log, suitable for process-mining techniques. If the desktop is captured during the recording process, information about enabled activities is available (next to the executed activity). This information can be extracted and added to an event log, resulting in a so-called translucent event log. A translucent event log can also be extracted from running information systems or created by applying domain knowledge to existing non-translucent event logs. Information about enabled activities is valuable for multiple reasons. For example, one may use this information to discover process models that capture the underlying behavior better. However, until now, only limited work has been done on exploiting the information on enabled activities for process discovery. In this work, we introduce translucent activity relationships derived from translucent event logs. These relationships can be embedded in various discovery algorithms. In this work, we focus on the Inductive Miner. In this process, we derive a translucent directly-follows graph. Based on this graph, we extend the Inductive Miner to use translucent activity relationships. We create three different variants and evaluate them on multiple translucent event logs by comparing them with the Inductive Miner. Based on the evaluation, we show that considering these relationships, fewer recordings are needed to discover a similar, if not even a superior, process model.