Visualizing the Process of Process Modeling with PPMCharts

In the quest for knowledge about how to make good process models, recent research focus is shifting from studying the quality of process models to studying the process of process modeling (often abbreviated as PPM) itself. This paper reports on our efforts to visualize this specific process in such a way that relevant characteristics of the modeling process can be observed graphically. By recording each modeling operation in a modeling process, one can build an event log that can be used as input for the PPMChart Analysis plug-in we implemented in ProM. The graphical representation this plug-in generates allows for the discovery of different patterns of the process of process modeling. It also provides different views on the process of process modeling (by configuring and filtering the charts).

💡 Research Summary

The paper addresses a gap in the process‑modeling literature by shifting the focus from the static quality of finished process models to the dynamics of how those models are created – the Process of Process Modeling (PPM). The authors argue that understanding the modeling activity itself can reveal systematic patterns, inefficiencies, and opportunities for improvement that are invisible when only the final artifact is examined. To capture this activity, they instrument a modeling environment so that every user operation—creation, deletion, connection, attribute change, and so on—is logged with a timestamp, operation type, and the identifier of the affected model element. The resulting event log, exported in the standard XES format, serves as input for a newly developed ProM plug‑in called PPMChart.

PPMChart visualizes the log as a two‑dimensional timeline. The horizontal axis represents elapsed time, while the vertical axis lists the individual model elements (activities, gateways, data objects, etc.). Each logged operation is rendered as a colored bar or dot; colors encode operation categories (e.g., add, modify, delete) and the length of a bar reflects the duration of the operation. The plug‑in is built on D3.js, providing interactive zooming, panning, and tool‑tips that reveal the full log entry when a visual element is clicked. Users can filter by operation type, by specific modeler, or by model file, and can reconfigure the chart to emphasize particular aspects of the modeling process.

The authors evaluated the approach with two empirical studies. In the first, 30 university students performed a BPMN modeling task while their actions were recorded. In the second, 12 professional modelers from industry completed a comparable task. For each participant, a PPMChart was generated and examined for recurring patterns. Several distinct behaviors emerged:

1. Explosive Insertion – a burst of element‑creation operations at the very beginning of the session, indicating a “big‑bang” design strategy.
2. Iterative Refinement – repeated modifications of the same gateway or connector, suggesting uncertainty about control‑flow semantics.
3. Spike‑Deletion – a short interval near the end of the session where a high frequency of delete‑and‑re‑add actions occurs; this spike correlates with lower post‑hoc model quality scores, implying that frantic last‑minute changes may degrade the final model.

The visual patterns also expose idle periods, concentrated work bursts, and the temporal ordering of design decisions, all of which can be linked to cognitive load and modeling expertise.

From a technical standpoint, the contribution lies in the seamless integration of fine‑grained modeling logs into the ProM process‑mining ecosystem, the definition of a compact event schema that abstracts away tool‑specific details, and the creation of an interactive, configurable chart that can be used both for research analysis and for pedagogical feedback. The authors acknowledge limitations: the current implementation supports a limited set of operation types, does not yet handle concurrent edits in collaborative settings, and relies on manual instrumentation of the modeling tool.

Future work is outlined along three axes: (1) extending the logging framework to automatically classify operations using machine‑learning techniques, (2) adapting the visualization to multi‑user collaborative logs, and (3) embedding the chart in a real‑time coaching system that alerts modelers when inefficient patterns (e.g., repeated deletions) arise, thereby supporting better modeling practices and higher‑quality process models.

In summary, the paper demonstrates that visualizing the process of process modeling with PPMChart provides actionable insights into modeling behavior, opens new avenues for empirical research on modeling cognition, and offers a practical tool for improving both education and professional practice in business process management.