Detecting Dynamic Relationships in Object-Centric Event Logs
Alessandro Gianola, Zeeshan Hameed, Marco Montali, Anjo Seidel, Mathias Weske, Sarah Winkler · 2026
Read Detecting Dynamic Relationships in Object-Centric Event Logs by Alessandro Gianola, Zeeshan Hameed, Marco Montali, Anjo Seidel, Mathias Weske, Sarah Winkler (2026) 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-2604-13053.txt.
Introduction
Gianola and five colleagues pin down how changing relationships between objects can be represented in object-centric event logs. Such logs leave object links underspecified, especially when links shift over time, muddying what the recorded process means. The team formally defines, for the first time, assumptions that make dynamic relationships unambiguous to store and manipulate, then checks existing logs and finds the assumptions often hold.
Catalog record
| CREATOR | Alessandro Gianola, Zeeshan Hameed, Marco Montali, Anjo Seidel, Mathias Weske, Sarah Winkler |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22867 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.13053
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Detecting Dynamic Relationships in Object-Centric Event Logs Alessandro Gianola, Zeeshan Hameed, Marco Montali, Anjo Seidel, Mathias Weske, Sarah Winkler Abstract Object-centric process mining examines how processes interact with multiple co-evolving objects, and has gained great interest in recent years. However, object-centric event logs (OCELs) leave object relationships underspecified in several respects, especially if relationships are dynamic, i.e., they change over time. In this paper, we identify and formally define for the first time assumptions that allow to represent and manipulate dynamic relationships in OCELs in a semantically unambiguous way. We evaluate existing logs to show that our assumptions are often satisfied, ensuring full transparency of relationship semantics. 1 Introduction Process mining aims to extract actionable insights from the execution data recorded by information systems [ 13 ] . By analyzing event logs, it seeks to reconstruct the actual behavior of operational processes and reason about performance, deviations, and conformance. While traditional process mining assumes that events refer to a single case notion, modern information systems store data about multiple interacting process instances that share co-evolving objects [ 14 ] . This has motivated the development of object-centric process mining, which studies processes that simultaneously evolve several interrelated objects of different types. Executions are recorded using the standard OCEL format (object-centric event logs [ 8 ] ), where each event may reference multiple objects. A variety of modeling formalisms has been proposed to represent object-centric processes: object-centric Petri nets (OCPN) [ 12 ] extend classical Petri nets to handle multiple object types. OCPNs are relatively simple and amenable to discovery, but known to be underspecified as they cannot track object identities and relationships [ 15 ] . More expressive formalisms capturing some form of synchronization include synchronous proclets [ 6 ] , object-centric process trees [ 16 ] , and object-centric Petri nets with identifiers [ 9 ] . While these formalisms can enforce synchronization semantics, they almost exclusively focus on stable relationships between objects that remain unmodified over time. The current literature thus almost entirely leaves out the important case where relationships dynamically change over time, despite the fact that such relationships are ubiquitous in real life processes: for example, a package might be moved from one truck to another, the destination address of a shipment can be dynamically updated, or the content of an order can change. This shortcoming in current research is largely due to the fact that OCEL, as well as the standard of object-centric event data (OCED) [ 5 ] , neither relate object relationships with events, nor with time. In this paper, we illustrate the resulting ambiguity by a number of examples. We then identify and formally define for the first time assumptions regarding how to log in an OCEL object-centric event data with dynamically changing relationships in an unambiguous way, with minimal additional domain knowledge. By evaluating available OCELs, we show that these assumptions are indeed respected to a great extent. 2 OCELs and Dynamic Relationships Contemporary object-centric event data representations [ 8 ] lack an explicitly defined semantics regarding the way events impact the evolution of object-to-object relationships— that is, which are the temporal intervals in which links between objects hold. For clarity, we use relationship at the type level, and link at the instance level. E.g., the type-level information that “an order contains items” constitutes a relationship, while the fact that “order o_{1} contains item i_{2} ” indicates a link. Such relationships are in the literature almost always assumed to be rigid (or stable ), that is, once established, a link never ceases to exist [ 3 , 7 , 11 ] . In this paper we call a dynamic many-to-one relationship between a many-type \mu and a one-type \sigma a relationship in which at every point in time, every object of type \mu is linked to at most one object of type \sigma . In contrast, in many-to-many relationships between types \mu and \sigma , an object of type \mu can at some point be linked to many objects of type \sigma and vice versa. In the sequel, many-to-one relationships are always supposed to be dynamic. The goal of this section is to provide a minimally intrusive mechanism to unambiguously view how an OCEL tracks the evolution of links of relationships. Here “minimally intrusive” means that we want to limit, as much as possible, the number of assumptions made, and the additional domain knowledge needed. For completeness, we recall the notion of object-centric event logs [ 12 ] : Definition 1 Given sets of object types \Sigma , activities \mathcal{A} , and timestamps \mathbb{T} , an object-centric event log (OCEL) is a tuple L=(E,\mathcal{O},\pi_{act},\pi_{obj},\pi_{time}) where: (i) E is a set of events; (ii) \mathcal{O} is a set of object identifiers typed by the function ot\colon\mathcal{O}\to\Sigma ; (iii) the functions \pi_{act}\colon E\to\mathcal{A} , \pi_{obj}\colon E\to\mathcal{P}(\mathcal{O}) , and \pi_{time}\colon E\to\mathbb{T} , mapping each event to an activity, a set of objects, and a timestamp. First and foremost, there are two possible ways of interpreting events in an OCEL, as illustrated by the following example. Example 1 Consider the situation where an order o_{1} is created with items i_{1} and i_{2} , then i_{2} gets removed from o_{1} , and later i_{3} is added. Depending on how object participation in events is interpreted, this can be expressed in two ways. In what we call the parameter semantics , objects denote the actual parameters of the activity referred by the event. An OCEL could thus look as follows: event activity order items \#_{1} \mathsf{create\ order} o_{1} i_{1},i_{2} \#_{2} \mathsf{reduce\ order} o_{1} i_{2} \#_{3} \mathsf{add\ to\ order} o_{1} i_{3} Note that each event is associated with the objects to which the respective activity applies. The current state after each event, in particular regarding which links are established resp. removed, remains however implicit. On the other hand, one can adopt a so-called snapshot semantics , where the objects listed in each event do not represent parameters, but provide instead a snapshot of the new state (e.g., the new state of an order) after the respective activity, making the existing links explicit. This would yield the following log: event activity order items \#_{1} \mathsf{create\ order} o_{1} i_{1},i_{2} \#_{2} \mathsf{reduce\ order} o_{1} i_{1} \#_{3} \mathsf{add\ to\ order} o_{1} i_{1},i_{3} Note how in event \#_{2} , the associated item is i_{1} , i.e., the remaining item in o_{1} – witnessing that the set of active links for the order-item relationship consists only of (o_{1},i_{1}) . This is in contrast to the table above, where event \#_{2} is associated with i_{2} , the item to which the removal operation is applied . When adopting parameter semantics, it is impossible to reconstruct from an OCEL alone which links hold at a given point in time: additional transactional domain knowledge on the updates semantics of activities is needed [ 4 ] . Indeed, even though this is left implicit, existing approaches typically adopt the snapshot semantics [ 7 ] . We hence follow this assumption. Assumption 1 Events are recorded under the snapshot semantics, that is, listing which objects are related as an effect of the event. However, the next example shows that even under the snapshot semantics, there is inherent ambiguity in