Semantic Process Discovery from User Interaction Logs

Processes structure the operations of organizations, from large-scale multinational to startup and from commercial enterprise to healthcare provider. To perform these processes, workers frequently interact with IT systems, such as ticketing systems or customer-relations management tools. These interactions leave digital breadcrumbs, capturing who performed which task, at what time, in which context, and with what outcome. The field of process mining focuses on the analysis of such breadcrumbs, captured in the form of event logs, to gain insights into organizational processes and identify improvement opportunities in organizations. However, traditional process mining has a blind spotit focuses on the analysis of system-generated backend events, such as an ERP system recording that an order was received or an HR system that shows that a contract was sent to an applicant. As a result, user interactions that do not result in such backend events or take place in other applications are excluded from consideration. As a result, important steps may be ignored, leading to an incomplete picture of organizational operations. Our research will bridge this gap by shifting the focus from system logs to user interaction (UI) logs. UI logs record every action users take on their screens, such as clicking, typing, or selecting options. These logs provide a more complete picture of how work gets done across various applications, including emails, spreadsheets, and web-based platforms. Therefore, analyzing UI logs can provide insights that were so far out of reach. However, working with UI logs comes with its own challenges. UI logs contain vast amounts of low-level data that need to be structured and refined before they can be useful. To address this, our research tackles two main challenges: The first challenge is transforming UI logs into structured process data. Since UI logs capture raw interactions without context, they must be processed to make sense of user activities. This involves assigning meaningful labels to actions, filtering out irrelevant data, and linking related interactions to ensure they are correctly grouped into process instances. The second challenge is turning this structured data into actionable business insights. Because UI logs capture highly detailed and granular events, simply applying traditional process mining techniques can yield overwhelmingly complex results. To address this, we must group low-level interactions into meaningful business activities, assign intuitive labels, and develop new visualization methods that present processes in a clear and insightful manner. By addressing these challenges, this project will significantly enhance how organizations analyze their processes, ensuring that organizations have a complete and accurate view of how work truly happens.