To bind or not to bind? Discovering Stable Relationships in Object-centric Processes (Extended Version)
Anjo Seidel, Sarah Winkler, Alessandro Gianola, Marco Montali, Mathias Weske · 2025
Read To bind or not to bind? Discovering Stable Relationships in Object-centric Processes (Extended Version) by Anjo Seidel, Sarah Winkler, Alessandro Gianola, Marco Montali, Mathias Weske (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-18231.txt.
Introduction
Anjo Seidel and co-authors work on process mining for business processes that involve many objects at once. Their paper appeared in 2025. Standard models mined from event logs show how orders, items, and machines flow and meet. They do not say which objects truly belong together, so they cannot flag runs that break intended relationships. Formal models with object identities can, but no one had shown how to discover them from logs.
The authors bridge the gap with explicit data models. They show that logs quietly assume stable many-to-one relationships between objects. Building those links into the mined model yields the formal model, with synchronization of related objects made exact. The mapping is proven, not just proposed. It gives process analysts a way to catch violations that simpler models wave through.
Catalog record
| CREATOR | Anjo Seidel, Sarah Winkler, Alessandro Gianola, Marco Montali, Mathias Weske |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 125 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-18231
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
To bind or not to bind? Discovering Stable Relationships in Object-centric Processes (Extended Version) Anjo Seidel, Sarah Winkler, Alessandro Gianola, Marco Montali, Mathias Weske Abstract Object-centric process mining investigates the intertwined behavior of multiple objects in business processes. From object-centric event logs, object-centric Petri nets (OCPN) can be discovered to replay the behavior of processes accessing different object types. Although they indicate how objects flow through the process and co-occur in events, OCPNs remain underspecified about the relationships of objects. Hence, they are not able to represent synchronization, i.e. executing objects only according to their intended relationships, and fail to identify violating executions. Existing formal modeling approaches, such as object-centric Petri nets with identifiers (OPID), represent object identities and relationships to synchronize them correctly. However, OPID discovery has not yet been studied. This paper uses explicit data models to bridge the gap between OCPNs and formal OPIDs. We identify the implicit assumptions of stable many-to-one relationships in object-centric event logs, which implies synchronization of related objects. To formally underpin this observation, we combine OCPNs with explicit stable many-to-one relationships in a rigorous mapping from OCPNs to OPIDs explicitly capturing the intended stable relationships and the synchronization of related objects. We prove that the original OCPNs and the resulting OPIDs coincide for those executions that satisfy the intended relationships. Moreover, we provide an implementation of the mapping from OCPN to OPID under stable relationships. 1 Introduction Process mining analyzes the behavior of information systems through event logs to gain insight into the behavior of the system [ 18 ] . Process mining techniques strive between two contrasting forces when dealing with process models: capturing complex real-life processes demands expressiveness , while performance and memory consumption of the adopted algorithms call for simplicity . Balancing these contrasting forces is difficult given that within a process mining pipeline, the same model can be seen as descriptive, like a model discovered from data, or as prescriptive, e.g., when later on used for conformance checking: while a descriptive model can favor simplicity over precision, a prescriptive one should be expressive enough to precisely capture the intended behavior. Achieving a suitable trade-off appears particularly difficult within the novel research area of object-centric process science, where processes co-evolve multiple, interrelated objects at once [ 16 ] . Executions of such processes are typically represented in object-centric event logs (OCELs) [ 6 ] , where each recorded event may refer to multiple objects. For example, consider the following OCEL L_{1} that records activities in a bicycle manufacturing workshop. For a given order, a frame, a set of wheels, and a handlebar are collected before being assembled: \displaystyle L_{1}=[ \displaystyle\langle{\mathsf{collect},\{f_{1},h_{1},w_{1},w_{2}\}}\rangle,\langle{\mathsf{assemble\_w},\{f_{1},w_{1},w_{2}\}}\rangle, \displaystyle\langle{\mathsf{assemble\_h},\{f_{1},h_{1}\}}\rangle,\langle{\mathsf{collect},\{f_{2},h_{2},w_{3},w_{4}\}}\rangle, \displaystyle\langle{\mathsf{assemble\_h},\{f_{2},h_{2}\}}\rangle,\langle{\mathsf{assemble\_w},\{f_{2},w_{3},w_{4}\}}\rangle \displaystyle] with f_{i} of type frame , h_{i} of type handlebar , and w_{i} of type wheel . Two bicycles are ordered; the first collect event takes the frame, handlebar, and two wheels for the first bike; the assemble_w event assembles two wheels to a frame, etc. To replay the event log and describe the behavior captured in an OCEL, an object-centric Petri net (OCPN) can be discovered [ 19 ] . For instance, the OCPN in 1 allows to replay the behavior of log L_{1} . In the initial marking ( ), all recorded objects in the log are placed, so that, e.g., all wheel objects w_{1},\ldots w_{4} reside initially in the wheel place. The transitions can fire by binding objects and processing them through the net. In OCPNs, a non-variable arc binds a single object, while a variable arc (double line) binds a set of objects. Finally, the net accepts a given execution if all objects are placed in the final marking ( ). Example 1 (OCPN for bicycle manufacturing) wheel frame handlebar \mathsf{collect} \mathsf{assemble\_w} \mathsf{assemble\_h} OCPNs favor simplicity over expressiveness: Although discovered OCPNs consider the token flow of multiple object types, OCPNs do not provide mechanisms to track the identity of objects, their relationships, and cardinality constraints. As pointed out in [ 17 ] , this deliberate underspecification makes discovered OCPNs underfitting, as they allow for too much behavior. This becomes apparent if the expected data model with cardinality constraints for log L_{1} is explicitly modeled as a UML class diagram: wheel frame handlebar 1..* 1..1 1..1 1..1 The behavior recorded in L_{1} satisfies the cardinality constraints for a many-to-one relationship between the wheels and a frame, and a one-to-one relationship between the frame and handlebar. However, in an OCPN it is impossible to express that a transition should bind objects in accordance with their mutual relationships, and it cannot distinguish transition firings that satisfy this criterion from those that bind unrelated objects, a distinction that is often essential when dealing with synchronization [ 8 ] . For example, in the OCPN in Example 1 wheels collected for a given frame may be reshuffled, that is, assembled with a different frame. Hence, it accepts the execution in the following log: \displaystyle L_{2}= \displaystyle[\langle{\mathsf{collect},\{f_{3},h_{3},w_{5},w_{6}\}}\rangle,\langle{\mathsf{collect},\{f_{4},h_{4},w_{7}\}}\rangle,\langle{\mathsf{assemble\_w},\{f_{3},w_{5}\}}\rangle, \displaystyle\langle{\mathsf{assemble\_h},\{f_{3},h_{4}\}}\rangle,\langle{\mathsf{assemble\_w},\{f_{4},w_{6},w_{7}\}}\rangle,\langle{\mathsf{assemble\_h},\{f_{4},h_{3}\}}\rangle] where wheel w_{6} is attached to frame f_{4} even though it was collected with frame f_{3} . This indicates that whenever an OCPN is used in a prescriptive manner for conformance checking or simulation, further so-called “bounding behavior” [ 17 ] is needed to ensure the desired synchronization. This bounding behavior can be suitably captured through more sophisticated and expressive object-centric process modeling languages, such as variants of Petri nets with identifiers (PNIDs) [ 7 , 20 ] , or synchronous proclets [ 4 ] . Recently, object-centric Petri nets with identifiers (OPIDs [ 8 ] ) have been proposed as a formalism that unifies the features of OCPNs, PNIDs, and synchronous proclets. In contrast to OCPNs, OPIDs are appropriate for prescriptive purposes because of their high expressiveness. However, while conformance for OPIDs is handled in [ 8 ] , the discovery of OPIDs and related formalisms has not yet been studied. To avoid reshuffling like in the above example, and similar synchronization problems, we start by identifying an important assumption that is implicit in OCELs: executions are supposed to satisfy stable many-to-one relationship between two object types which synchronize the corresponding objects throughout the execution. This means that if the relationship between object types \sigma_{m} and \sigma_{o} is many-to-one, then whenever an object of type \sigma_{m} is bound to an object of type \sigma_{o} , it is never bound to a different object of type \sigma_{o} . This implicit assumption comes from the fact that OCEL assumes global and rigid relationships throughout execution [ 10 , 2 ] . To find a model that satisfies both the simplicity of discovered OCPNs and the expressiveness to enforce the desired bounding behavior, this paper presents a rigorous mapping from an OCPN together with a set of stable many-to-one relationships, to an equivalent replay OPID that enforces the intended synchronization. By doing so, we at once provide a fourfold contribution: (1) we provide a first discovery technique for one specific class of OPIDs, (2) we identify the class of relatively simple replay OPIDs that are yet expressive enough to enforce the desired bounding behaviour, (3) we formally prove that a so-obtained replay OPID is equivalent to the original OCPN for executions that respect the data model (4) we provide a proof-of-concept implementation