High-Level Event Mining: Overview and Future Work

TL;DR

This work tackles the limitation of traditional process mining in capturing system-level dynamics by introducing high-level event mining. It formalizes high-level events as observations derived from process components (activities, resources, segments) and process aspects across fixed time windows, with thresholds determining when events emerge. It then develops propagation mechanisms through proximity functions to connect high-level events into cascades and threads, and introduces the interplay with underlying instances to relate patterns to case properties. A robustness framework is proposed to study disruptions and their effects on waiting times, and future directions include online predictions, adaptive framing, and interactive tools. Overall, the approach enables analysis of congestion, load propagation, and resilience in complex processes beyond what low-level events reveal.

Abstract

Process mining traditionally relies on input consisting of low-level events that capture individual activities, such as filling out a form or processing a product. However, many of the complex problems inherent in processes, such as bottlenecks and compliance issues, extend beyond the scope of individual events and process instances. Consider congestion, for instance, it can involve and impact numerous cases, much like how a traffic jam affects many cars simultaneously. High-level event mining seeks to address such phenomena using the regular event data available. This report offers an extensive and comprehensive overview at existing work and challenges encountered when lifting the perspective from individual events and cases to system-level events.

Figures (15)

• Figure 1: The citizenship application process involves initial submission, review, optional updates, and decision-making steps at the foreign office. Jane manages submissions and updates, while Mike and Sarah review and decide whether the applications will be approved or denied
• Figure 2: The dark dots (left) correspond to events from set $f_{\mathit{exec}}^{\mathit{ev}}(\mathit{review},w)$ which indicate executions of activity review during window $w$. The dark dots (right) correspond to events from set $f_{\mathit{enqueue}}^{\mathit{ev}}(\mathit{review},w))$ which indicate cases enqueuing for activity review during window $w$.
• Figure 3: The dark dots correspond to events from set $f_{\mathit{queue}}^{\mathit{ev}}(\mathit{review},w)$ which indicate cases in the queue for activity review that have entered the queue before or during window $w$.
• Figure 4: The dots encircled in red in the left illustration represent events that were executed by Sarah during window $w$ (set $f_{\mathit{do}}^{\mathit{ev}}(\mathit{Sarah},w)$). The dark dots in the right illustration correspond to a subset of the events occurring during $w$ (specifically review events) whose next task (activity approve or deny) will be handled by Sarah (set $f_{\mathit{todo}}^{\mathit{ev}}(\mathit{Sarah},w)$).
• Figure 5: The dots encircled in red represent events that were executed by Sarah. The dark dots correspond to a subset of the events occurring before or during $w$ (specifically review events) whose next task (activity approve or deny) will be handled by Sarah (set $f_{\mathit{workload}}^{\mathit{ev}}(\mathit{Sarah},w)$).

Theorems & Definitions (18)

• definition thmcounterdefinition: Events, Event log
• definition thmcounterdefinition: Traces, Steps, Segments
• definition thmcounterdefinition: Previous and Next event
• definition thmcounterdefinition: Framing, Time Windows
• definition thmcounterdefinition: Process components
• definition thmcounterdefinition: Aspect functions
• definition thmcounterdefinition: High-level event
• definition thmcounterdefinition: High-level activity
• definition thmcounterdefinition: Proximity function
• definition thmcounterdefinition: Instance overlap