Discovering Two-Level Business Process Models from User Interface Event Logs

Abstract

The widespread adoption of Robotic Process Automation (RPA) to automate repetitive tasks has surged, utilizing front-end interfaces to replicate human actions. Task mining is integral to the RPA lifecycle, capturing and analyzing fine-grained user interactions. As opposed to traditional process mining, which focuses on business processes, task mining relies on low-level events from user interface (UI) logs. Bridging the gap between task mining and process mining, this paper introduces a novel approach to generate two-level process models from UI logs. This method addresses challenges related to granularity and screen ambiguities by considering two levels of abstraction. It identifies each activity at the screen level for a comprehensive overview and delves into finegrained user actions for a deeper understanding. This dual-level model aids in resolving ambiguities in inferred user intention, offering labeled activities for enhanced understandability. The proposed method was validated using synthetically generated UI event logs for an example with real-world characteristics. The research highlights the importance of contextual information in UI event logs and provides a solution to effectively map low-level UI interactions to higher-level user activities for meaningful analysis and automation.