Goal-driven Variant Categorization

Daniel Calegari; Daniel Amyot

cs.AISep 18, 2026

Goal-driven Variant Categorization

Authors: Daniel Calegari, Daniel Amyot

Organizations: Universidad ORT Uruguay, Montevideo, Uruguay · University of Ottawa, Ottawa, Canada

Abstract

Process discovery rarely yields a single coherent process structure. For analysis, a common step is to cluster process variants based on structural similarity and then assign business meaning to the resulting groups. Since these partitions are not derived from the organization's goals, analysts must manually interpret and consolidate variants into business-meaningful categories. This judgment-intensive step becomes increasingly difficult as the number and complexity of variants grow. In this paper, we propose a goal-driven approach to variant categorization that reverses this workflow. We first author an organization's goal model that predefines the categorization axis. Each variant is transformed into a textual narrative describing its behavior, and a Large Language Model (LLM) interprets it in the context of the goal model and assigns the variant to the most appropriate category. LLM-based semantic reasoning connects low-level process behavior with analyst-defined business goals. We instantiate this approach end-to-end and evaluate it on three public logs differing substantially in scale and behavioral diversity. Goal-model guidance yields partitions that differ from those produced by unguided induction and respond to controlled edits to the declared alternatives, at the cost of authoring a goal model.

Figures & tables

Panel of Figure 1 — Figure 1. The RTFM goal model in GRL notation: rounded rectangles are goals, hexagons are tasks, the barred hexagon is a KPI, and the curved outlines are softgoals. Decomposition links carry a filled head at the decomposed parent and its And/Or/Xor operator; contribution links carry an open head and their negative/positive weight in the $[-100,100]$ interval. Goal model for the Road Traffic Fine Management process in GRL notation. One actor, the issuing police back-office, holds a goal tree AND-decomposed at the top into issuing a fine and resolving it, and OR/XOR-decomposed below into five ways a case can end: timely payment, delinquent payment, administrative appeal, judicial appeal, and coercive credit collection. Administrative appeal is AND-decomposed into four mandatory sub-steps. Three softgoals describe organizational priorities, and one KPI converts duration into a satisfaction level.

Panel of Figure 2 — Figure 2. The GoalCat pipeline (phases I–IV). Solid arrows are sequence flows; dashed arrows are data associations, not step transitions. The \GoalCat{} pipeline has four phases. Phase I extracts, profiles, textualizes, and samples trace variants. Phase II induces a taxonomy through either intent-guided or open induction. Phase III assigns every variant to a category or to the residual and analyzes the resulting categories. Phase IV supports analyst review and may return to taxonomy induction or goal-model revision.

ID	Narrative	Guided category
V0004	Create Fine Payment (+1d)	timely payment
V0003	… Add penalty (+60d) Payment (+344d)	delinquent payment
V0007	… Add penalty (+60d) Payment (+10d) Payment (+30d) …	delinquent payment
V0008	… Payment (+45d) Add penalty (+15d) Payment (+10d) …	delinquent payment
V0002	… Add penalty (+60d) Send for Credit Collection (+745d)	coercive collection
V0005	… Add penalty (+60d) Appeal to Judge (+21d)	judicial appeal

Table 1. Excerpt of RTFM narratives and the guided-arm category from the initial induction.

Log	Axis	Variants	Cases	$\bm{\|T^{G}\|}$	$\bm{n}$	Residual (var.)	Residual (cases)
RTFM	resolution	231	150,370	5	5	2.8% [0.9–7.4]	13.8%
Sepsis	admission	846	1,050	2	5	10.4%	22.9%
Sepsis	discharge	846	1,050	5	5	18.2% [17.1–21.0]	29.6% [28.6–32.1]
BPIC 2019	matching regime	11,973	251,734	4	3	20.7% [19.3–21.6]	16.3% [16.0–16.8]

Table 2. Evaluation logs and the intent-guided partition (

T^{G}

		Variant-weighted			Case-weighted
Log	Axis	Guided-guided	Open-open	Guided vs. open	Guided-guided	Open-open	Guided vs. open
RTFM	resolution	0.58 [0.49–0.69]	0.50 [0.42–0.65]	0.26 [0.15–0.46]	0.99 [0.98–1.00]	0.93 [0.87–1.00]	0.91 [0.87–0.96]
Sepsis	admission	0.88 [0.84–0.92]	0.51 [0.43–0.56]	0.30 [0.27–0.32]	0.92 [0.89–0.94]	0.61 [0.55–0.66]	0.44 [0.40–0.46]
Sepsis	discharge	0.76 [0.68–0.86]	0.51 [0.43–0.56]	0.25 [0.21–0.32]	0.79 [0.72–0.88]	0.61 [0.55–0.66]	0.36 [0.30–0.43]
BPIC 2019	match. regime	0.51 [0.48–0.54]	0.40 [0.36–0.43]	0.17 [0.17–0.18]	0.69 [0.61–0.83]	0.54 [0.45–0.70]	0.40 [0.27–0.49]

Table 3. Partition agreement (AMI, residual as its own block), mean [min–max] over replicate pairs. Each arm’s own agreement is the noise floor for the guided-vs-open column.

Log	Axis	Edit	Target rate ( $\bm{n}$ )	Collateral rate ( $\bm{n}$ )	$\bm{C^{\emptyset}}$ range
RTFM	resolution	A: remove coercive collection	100.0% (34)	20.5% [17.8–24.9] (197)	15.7–39.6%
RTFM	resolution	B: merge delinquent + coercive	97.5% [93.8–100] (81)	20.7% [16.7–23.3] (150)	16.0–43.3%
RTFM	resolution	C: add fine annulment †	0.0% [0.0–0.0] uptake (n/a)	19.7% [19.0–21.6] (231)	14.3–35.1%
Sepsis	admission	A: remove Admission IC	100.0% (89)	0.05% [0.00–0.13] (757)	0.4–2.4%
Sepsis	admission	C: add High-Dependency Unit †	0.0% [0.0–0.0] uptake (n/a)	2.2% [1.4–3.2] (846)	1.2–2.7%
Sepsis	discharge	A: remove Release E	100.0% (6)	8.9% [6.9–11.7] (840)	3.2–9.3%

Table 4. Goal-model perturbations, guided arm, mean [min–max] over five replicates per condition (Sepsis admission has no merge: OR point with only two children).

n

in the Target column is the number of variants realizing the perturbed alternative(s) (n/a for C);

n

in the Collateral column is the rest. † C’s Target column reports DistractorUptake in place of TargetReassignment.

Process Mining

Process mining turns event logs of real-world activities into models and predictions for understanding, checking, and improving workflows.

See all 45 papers

Framework for Grouping Local Process Models
Jul 6, 2026Viki Peeva, Wil M. P. van der AalstProcess Mining Clustering
Chair of Process and Data Science (PADS), RWTH Aachen University, Ahronstrasse 55, Aachen, 52074, North Rhine-Westphalia, Germany.
Pragmos: A Process Agentic Modeling System
Apr 30, 2026Pedro-Aarón Hernández-Ávalos, Luciano García-BañuelosAgentic Workflows Business Process Modeling
Tecnologico de Monterrey, Mexico
Evaluating LLM-Based Goal Extraction in Requirements Engineering: Prompting Strategies and Their Limitations
Apr 24, 2026Anna Arnaudo, Riccardo Coppola, Maurizio Morisio +4Requirements Engineering LLM Prompting
Department of Control and Computer Engineering, Politecnico di Torino · Politecnico di Torino

Jul 6, 2026J/K move · Enter open · S save

Framework for Grouping Local Process Models

Viki Peeva, Wil M. P. van der Aalst

Chair of Process and Data Science (PADS), RWTH Aachen University, Ahronstrasse 55, Aachen, 52074, North Rhine-Westphalia, Germany.

Process Mining Clustering

Open paper AskPDF arXiv

Local Process Models (LPMs) are an underexplored concept in process mining. LPMs describe patterns in event data considering sequence, choice, concurrency, and loop. In recent years, process mining has proved successful in the analysis and improvement of operational processes. More often than not, surprising findings are found when one does not consider the full process, making LPMs and their discovery highly valuable. However, similar to other pattern mining approaches, LPM discovery algorithms face the problems of model explosion and model repetition, i.e., the algorithms may create hundreds if not thousands of LPMs, and subsets of them are close in structure or behavior. Practically, no analyst would be able to comb through thousands of LPMs leading to using a sample of LPMs that are easily accessible. The current sentiment is that the top-scoring LPMs form the optimal sample to be presented. However, different applications should demand a different optimal sample. With this work, we show that if the goal of the mined LPMs is to understand a process, using the top-scoring LPMs as an optimal sample is a poor choice because of high repetition. We propose a framework for grouping LPMs and creating an optimal sample by taking one representative LPM for each group. We measure similarity between models via established process model similarity measures or by comparing the context in which an LPM appears. The context is formed using data attributes available in the underlying event logs. We demonstrate the usefulness of grouping on multiple event logs by comparing repetition and coverage between samples comprised of the top-scoring models and the representatives of discovered groups.

Abstract

Process Mining

Explore similar work

Framework for Grouping Local Process Models

Pragmos: A Process Agentic Modeling System

Evaluating LLM-Based Goal Extraction in Requirements Engineering: Prompting Strategies and Their Limitations