Process discovery often yields multiple inconsistent process structures. Analysts typically cluster variants by structural similarity and then manually assign business meaning, but these partitions are not derived from organizational goals, requiring extensive judgment to merge variants into meaningful categories. As the number and complexity of variants grow, this step becomes increasingly difficult.
We propose a goal‑driven variant categorization approach that reverses the traditional workflow. First, the organization authors a goal model that defines the categorization axes. Each variant is then transformed into a textual narrative describing its behavior, and a Large Language Model (LLM) performs semantic reasoning within the context of the goal model to assign the variant to the most appropriate category. This LLM‑based reasoning bridges low‑level process behavior with analyst‑defined business goals.
We built an end‑to‑end system and evaluated it on three public logs that differ substantially in scale and behavioral diversity. Results show that goal‑model guidance produces partitions distinct from unguided induction and responds to controlled edits of declared alternatives, with the only cost being the effort to author a goal model.
Review: The approach injects business intent directly into the variant grouping process, markedly reducing manual interpretation effort and fitting scenarios with many variants and clear organizational goals.