Business process management employs process models and event logs to represent the behavior of the information systems under study. Traditional case-centric notions consider the order of activities and events in isolated process instances. The emerging field of object-centric processes challenges this assumption by putting objects in the center. Object-centric process mining and modeling approaches identify the structure of co-evolving data objects that influence the behavior of an information system to provide a comprehensive view of the system behavior. Object-centricity has been investigated independently in process modeling and in process mining, which resulted in the coexistence of seemingly contradictory assumptions and definitions. As a community effort, this research manifesto relates and aligns existing terminologies, definitions, and perspectives to provide a common ground for current and future research in object-centric business process management. Based on the current state of research, we propose a conceptualization that sets process models and event logs in relation to the information system’s behavior and the execution data it generates. The conceptualization aims at aligning different terminologies and, thus, providing a basis to model and analyze behavioral characteristics. Building on this common ground, we identify open research challenges along the most relevant research areas in object-centric process management. For each research area, its current status is investigated and an outline of the most relevant research challenges is presented.

Large Language Models (LLMs) are trained on a vast amount of text to interpret and generate human-like textual content. They are becoming a vital vehicle in realizing the vision of the autonomous enterprise, with organizations today actively adopting LLMs to automate many aspects of their operations. LLMs are likely to play a prominent role in future AI-augmented business process management systems (ABPMSs) catering functionalities across all system lifecycle stages. One such system's functionality is Situation-Aware eXplainability (SAX), which relates to generating causally sound and yet human-interpretable explanations that take into account the process context in which the explained condition occurred. In this paper, we present the SAX4BPM framework developed to generate SAX explanations. The SAX4BPM suite consists of a set of services and a central knowledge repository. The functionality of these services is to elicit the various knowledge ingredients that underlie SAX explanations. A key innovative component among these ingredients is the causal process execution view. In this work, we integrate the framework with an LLM to leverage its power to synthesize the various input ingredients for the sake of improved SAX explanations. Since the use of LLMs for SAX is also accompanied by a certain degree of doubt related to its capacity to adequately fulfill SAX along with its tendency for hallucination and lack of inherent capacity to reason, we pursued a methodological evaluation of the perceived quality of the generated explanations. To this aim, we developed a designated scale and conducted a rigorous user study. Our findings show that the input presented to the LLMs aided with the guard-railing of its performance, yielding SAX explanations having better-perceived fidelity. This improvement is moderated by the perception of trust and curiosity. More so, this improvement comes at the cost of the perceived interpretability of the explanation.

Object-centric process mining views processes and traces as an interaction between many objects, each with their own life cycle, as opposed to being centred around the concept of a single case. Instead of describing implicit process flows, declarative process modelling focuses on the description of processes as a set of explicit rules or constraints. The declarative Dynamic Condition Response (DCR) Graphs notation has seen wide industry adoption, in particular in the Danish public sector, has seen significant work on the development of methods for the modelling, verification, and enactment of collaborative processes, and has led to the development of the award winning DisCoveR process miner. In this paper we apply object-centric concepts to DCR Graphs modelling and mining, in particular we: (1) show an extension to DCR Graphs that allows capturing of object-centric process relations and (2) introduce a process discovery method for such object-centric DCR Graphs. We showcase these contributions on the BPIC2017 loan application log.

Process performance analysis based on event logs is a core task of process mining. Typical tools enrich a directly-follows graph with statistics on waiting times between activities. Such projection may reveal process issues that manifest as a high average waiting time between activities. However, the purely control-flow-oriented view disregards the influence of actor behavior on process performance and may lead to a distorted analysis. Typically, projected measures aggregate the waiting time it takes for disparate types of actor behavior to a single measure: a direct continuation of the work by the same actor, a continuation of the work by the same actor after being interrupted by another case, or a handover to another actor. For a handover, the receiving actor may decide to prioritize activities in other cases before starting the work. Hence, two similar waiting time measures may imply very different dynamics of the actors’ behavior. The paper contributes a method to systematically decompose the regular control-flow performance measure into more fine-grained performance measures based on such behavioral mechanisms of actors. We leverage event knowledge graphs as a joint representation of actor and control flow perspectives to derive features for the behavioral mechanisms and systematically analyze them. The evaluation of the features on a loan application process shows that they provide clearly interpretable performance insight compared to the potentially misleading average waiting times.

Maintaining a software system is a continuous and complex process, typically following a workflow defined by the responsible organization. However, in practice, developers often deviate from the defined process due to personal preferences, varying customer requirements, or urgent deadlines. Such deviations may cause problems later on, or they may indicate potential process improvements. To deal with deviations and improve processes, it is first necessary to fully understand the processes actually employed by developers. For this purpose, process-mining techniques have been proposed that primarily build on version-control data. In this paper, we present a complementary process-mining technique that uses Jira issues to recover process activities not visible in version-control data, particularly focusing on developers’ interactions with issues and each other. We conducted a case study with 74 repositories of 24 developer teams from one large international company to understand the technique’s merits. Our technique revealed process differences across teams and depending on the types of Jira issues, providing novel insights for the company that helped to better understand the employed processes.

Automatically discovering a process model from an event log is the prime problem in process mining. This task is so far approached as an unsupervised learning problem through graph synthesis algorithms. Algorithmic design decisions and heuristics allow for efficiently finding models in a reduced search space. However, design decisions and heuristics are derived from assumptions about how a given behavioral description — an event log — translates into a process model and were not learned from actual models which introduce biases in the solutions. In this paper, we explore the problem of supervised learning of a process discovery technique. We introduce a technique for training an ML-based model using graph convolutional neural networks, which translates a given input event log into a sound Petri net. We show that training this model on synthetically generated pairs of input logs and output models allows it to translate previously unseen synthetic and several real-life event logs into sound, arbitrarily structured models of comparable accuracy and simplicity as existing state of the art techniques in imperative mining. We analyze the limitations of the proposed technique and outline alleys for future work.

Numerous process discovery techniques exist for generating process models that describe recorded executions of business processes. The models are meant to generalize executions into human-understandable modeling patterns, notably parallelism, and enable rigorous analysis of process deviations. However, well-defined models with parallelism returned by existing techniques are often too complex or generalize the recorded behavior too strongly to be trusted in a practical business context. We bridge this gap by introducing the Probabilistic Inductive Miner (PIM) based on the Inductive Miner framework. PIM compares in each step the most probable operators and structures based on frequency information in the data, which results in block-structured models with significantly higher accuracy. All design choices in PIM are based on business context requirements obtained through a user study with industrial process mining experts. PIM is evaluated quantitatively and in an novel kind of empirical study comparing users’ trust in discovered model structures. The evaluations show that PIM strikes a unique trade-off between model accuracy and model complexity, that is conclusively preferred by users over all state-of-the-art process discovery methods.

Automatically discovering a process model from an event log is the prime problem in process mining. This task is so far approached as an unsupervised learning problem through graph synthesis algorithms. Algorithmic design decisions and heuristics allow for efficiently finding models in a reduced search space. However, design decisions and heuristics are derived from assumptions about how a given behavioral description – an event log – translates into a process model and were not learned from actual models which introduce biases in the solutions. In this paper, we explore the problem of supervised learning of a process discovery technique d. We introduce a technique for training an ML-based model d using graph convolutional neural networks; d translates a given input event log into a sound Petri net. We show that training d on synthetically generated pairs of input logs and output models allows d to translate previously unseen synthetic and several real-life event logs into sound, arbitrarily structured models of comparable accuracy and simplicity as existing state of the art techniques in imperative mining. We analyze the limitations of the proposed technique and outline alleys for future work.

To identify the causes of performance problems or to predict process behavior, it is essential to have correct and complete event data. This is particularly important for distributed systems with shared resources, e.g., one case can block another case competing for the same machine, leading to inter-case dependencies in performance. However, due to a variety of reasons, real-life systems often record only a subset of all events taking place. To understand and analyze the behavior and performance of processes with shared resources, we aim to reconstruct bounds for timestamps of events in a case that must have happened but were not recorded by inference over events in other cases in the system. We formulate and solve the problem by systematically introducing multi-entity concepts in event logs and process models. We introduce a partial-order based model of a multi-entity event log and a corresponding compositional model for multi-entity processes. We define PQR-systems as a special class of multi-entity processes with shared resources and queues. We then study the problem of inferring from an incomplete event log unobserved events and their timestamps that are globally consistent with a PQR-system. We solve the problem by reconstructing unobserved traces of resources and queues according to the PQR-model and derive bounds for their timestamps using a linear program. While the problem is illustrated for material handling systems like baggage handling systems in airports, the approach can be applied to other settings where recording is incomplete. The ideas have been implemented in ProM and were evaluated using both synthetic and real-life event logs. Comment: Final formatted version at Fundamenta Informatica

Remaining time prediction (RTP) is the problem of predicting the time until a specific process step is reached in a specific process instance. Feature engineering in established RTP techniques assume that cases progress in isolation. Intercase dynamics such as batching violate this assumption, leading to high prediction errors. Yet, existing RTP techniques do not consider the nature of prediction errors to improve quality. We contribute a technique for identifying the location and context of prediction errors by visually comparing prediction and ground truth. For the case of batching, we show how to engineer inter-case features that detail the impact of batching on the remaining time. Our evaluation shows that adding intercase features improves prediction performance across almost all evaluated primary prediction methods on two real-life event logs, with error reductions of up to 37%. We finally advocate for a more thorough and transparent evaluation of prediction errors in RTP research, including our own results.

Dynamic bottlenecks occur when some cases in a particular part of the process are temporarily delayed. In performance-optimized systems such as production systems, warehouse automation systems, and baggage handling systems, such bottlenecks are rare, bounded in time and location, but costly when they occur and propagate through the system. Detecting and understanding the situations that cause such bottlenecks is crucial for mitigating and preventing processing delays. Classical process mining techniques that analyze performance along individual cases cannot detect these phenomena and their causes. We show that undesired system-level behavior can be detected when identifying temporal event patterns across different cases in the same process step. Conceptualizing these patterns as system-level events allows us to correlate them into cascades of system-level behavior using spatio-temporal conditions. We discover classes of frequent patterns in these cascades that describe behaviors that precede bottlenecks. Applied on event data of a major European airport, our approach could fully automatically detect cascades of undesired system-level behavior leading to dynamic bottlenecks. Each detected cascade was verified as a correct causal explanation for a dynamic bottleneck due to the physical system layout and its processing.

Given a model of the expected behavior of a business process and given an event log recording its observed behavior, the problem of business process conformance checking is that of identifying and describing the differences between the process model and the event log. A desirable feature of a conformance checking technique is that it should identify a minimal yet complete set of differences. Existing conformance checking techniques that fulfill this property exhibit limited scalability when confronted to large and complex process models and event logs. One reason for this limitation is that existing techniques compare each execution trace in the log against the process model separately, without reusing computations made for one trace when processing subsequent traces. Yet, the execution traces of a business process typically share common fragments (e.g. prefixes and suffixes). A second reason is that these techniques do not integrate mechanisms to tackle the combinatorial state explosion inherent to process models with high levels of concurrency. This paper presents two techniques that address these sources of inefficiency. The first technique starts by transforming the process model and the event log into two automata. These automata are then compared based on a synchronized product, which is computed using an A* heuristic with an admissible heuristic function, thus guaranteeing that the resulting synchronized product captures all differences and is minimal in size. The synchronized product is then used to extract optimal (minimal-length) alignments between each trace of the log and the closest corresponding trace of the model. By representing the event log as a single automaton, this technique allows computations for shared prefixes and suffixes to be made only once. The second technique decomposes the process model into a set of automata, known as S-components, such that the product of these automata is equal to the automaton of the whole process model. A product automaton is computed for each S-component separately. The resulting product automata are then recomposed into a single product automaton capturing all the differences between the process model and the event log, but without minimality guarantees. An empirical evaluation using 40 real-life event logs shows that, used in tandem, the proposed techniques outperform state-of-the-art baselines in terms of execution times in a vast majority of cases, with improvements ranging from several-fold to one order of magnitude. Moreover, the decomposition-based technique leads to optimal trace alignments for the vast majority of datasets and close to optimal alignments for the remaining ones.

Predictive performance analysis is crucial for supporting operational processes. Prediction is challenging when cases are not isolated but influence each other by competing for resources (spaces, machines, operators). The so-called performance spectrum maps a variety of performance-related measures within and across cases over time. We propose a novel prediction approach that uses the performance spectrum for feature selection and extraction to pose machine learning problems used for performance prediction in non-isolated cases. Although the approach is general, we focus on material handling systems as a primary example. We report on a feasibility study conducted for the material handling systems of a major European airport. The results show that the use of the performance spectrum enables much better predictions than baseline approaches.

Considerable amounts of data, including process events, are collected and stored by organisations nowadays. Discovering a process model from such event data and verification of the quality of discovered models are important steps in process mining. Many discovery techniques have been proposed, but none of them combines scalability with strong quality guarantees. We would like such techniques to handle billions of events or thousands of activities, to produce sound models (without deadlocks and other anomalies), and to guarantee that the underlying process can be rediscovered when sufficient information is available. In this paper, we introduce a framework for process discovery that ensures these properties while passing over the log only once and introduce three algorithms using the framework. To measure the quality of discovered models for such large logs, we introduce a model-model and model-log comparison framework that applies a divide-and-conquer strategy to measure recall, fitness, and precision. We experimentally show that these discovery and measuring techniques sacrifice little compared to other algorithms, while gaining the ability to cope with event logs of 100,000,000 traces and processes of 10,000 activities on a standard computer.

The Business Process Management field addresses design, improvement, management, support, and execution of business processes. In doing so, we argue that it focuses more on developing modeling notations and process design approaches than on the needs and preferences of the individual who is modeling (i.e., the user). New data-centric process modeling approaches are taken as a relevant and timely stream of process design approaches to test our argument. First, we provide a review of existing data-centric process approaches, culminating in a theoretical classification framework. Next, we empirically evaluate three specific approaches with regard to the claims they make. We had participants representative of actual users try out these approaches on realistic scenarios via a series of workshops. Participants assessed to what extent quality claims from the literature could be recognized within the workshop sessions. The results of this evaluation substantiate a number of claims behind the approaches, but also identify opportunities to further improve them. Most prominently, we found that the usability aspects of all considered approaches are a source of concern. This leads us to the insight that usability aspects of process design approaches are crucial and, in the perception of groups representative of actual users, leave much to be desired. In that sense, our research can be seen as a wake-up call for process modeling notation designers to consider the usability side—and as such, the interest of the human modeler—more than is currently the case.

Artifact-centric modeling is an approach for capturing business processes in terms of so-called business artifacts — key entities driving a company's operations and whose lifecycles and interactions define an overall business process. This approach has been shown to be especially suitable in the context of processes where one-to-many or many-to-many relations exist between the entities involved in the process. As a contribution towards building up a body of methods to support artifact-centric modeling, this article presents a method for automated discovery of artifact-centric process models starting from logs consisting of flat collections of event records. We decompose the problem in such a way that a wide range of existing (non-artifact-centric) automated process discovery methods can be reused in a flexible manner. The presented methods are implemented as a package for ProM, a generic open-source framework for process mining. The methods have been applied to reverse-engineer an artifact-centric process model starting from logs of a real-life business process.

Business process models are an important means to design, analyze, implement, and control business processes. As with every type of conceptual model, a business process model has to meet certain syntactic, semantic, and pragmatic quality requirements to be of value. For many years, such quality aspects were investigated by centering on the properties of the model artifact itself. Only recently, the process of model creation is considered as a factor that influences the resulting model’s quality. Our work contributes to this stream of research and presents an explorative analysis of the process of process modeling (PPM). We report on two large-scale modeling sessions involving 115 students. In these sessions, the act of model creation, i.e., the PPM, was automatically recorded. We conducted a cluster analysis on this data and identified three distinct styles of modeling. Further, we investigated how both task- and modeler-specific factors influence particular aspects of those modeling styles. Based thereupon, we propose a model that captures our insights. It lays the foundations for future research that may unveil how high-quality process models can be established through better modeling support and modeling instruction.

Enterprise Resource Planning (ERP) systems are widely used to manage business documents along a business processes and allow very detailed recording of event data of past process executions and involved documents. This recorded event data is the basis for auditing and detecting unusual flows. Process mining techniques can analyze event data of processes stored in linear event logs to discover a process model that reveals unusual executions. Existing approaches to obtain linear event logs from ERP data require a single case identifier to which all behavior can be related. However, in ERP systems processes such as Order to Cash operate on multiple interrelated business objects, each having their own case identifier, their own behavior, and interact with each other. Forcing these into a single case creates ambiguous dependencies caused by data convergence and divergence which obscures unusual flows in the resulting process model. In this paper, we present a new semi-automatic, end-to-end approach for analyzing event data in a plain database of an ERP system for unusual executions. More precisely, we identify an artifact-centric process model describing the business objects, their life-cycles, and how the various objects interact along their life-cycles. This way, we prevent data divergence and convergence. We report on two case studies where our approach allowed to successfully analyze processes of ERP systems and reliably revealed unusual flows later confirmed by domain experts.

There is a continuous pressure to make healthcare processes more efficient and effective without sacrificing quality. Conformance checking can be used to improve processes by analyzing event data and directly relating observed behavior and modeled behavior. Conformance checking provides diagnostics that go far beyond measuring traditional key performance indicators. However, current conformance checking techniques focus on a rather simplistic setting where executions of process instances are sequential and homogeneous whereas healthcare processes are known to be dynamic, complex, and ad-hoc. In healthcare process instances of patients often follow a unique path through the process with one-of-a-kind deviations. Moreover, timestamps are often rather coarse (the date is known, but not the time) resulting in an unreliable ordering of events. As current techniques are unable to handle concurrent events, and the obtained sequential alignments are unable to provide structural information about deviations, the diagnostics provided are often insufficient and misleading. This paper presents a novel approach using partially ordered traces and partially ordered alignments which aims to incorporate unreliability and concurrency in the input while providing diagnostics about deviations that take causalities into account. The approach has been implemented in ProM and was evaluated using event data from a Dutch hospital.

This article addresses the transformation of a process model with an arbitrary topology into an equivalent structured process model. In particular, this article studies the subclass of process models that have no equivalent well-structured representation but which, nevertheless, can be partially structured into their maximally-structured representation. The transformations are performed under a behavioral equivalence notion that preserves the observed concurrency of tasks in equivalent process models. The article gives a full characterization of the subclass of acyclic process models that have no equivalent well-structured representation, but do have an equivalent maximally-structured one, as well as proposes a complete structuring method. Together with our previous results, this article completes the solution of the process model structuring problem for the class of acyclic process models.

Specification mining extracts candidate specification from existing systems, to be used for downstream tasks such as testing and verification. Specifically, we are interested in the extraction of behavior models from execution traces. In this paper we introduce mining of branching-time scenarios in the form of existential, conditional Live Sequence Charts, using a statistical data-mining algorithm. We show the power of branching scenarios to reveal alternative scenario-based behaviors, which could not be mined by previous approaches. The work contrasts and complements previous works on mining linear-time scenarios. An implementation and evaluation over execution trace sets recorded from several real-world applications shows the unique contribution of mining branching-time scenarios to the state-of-the-art in specification mining.

Live sequence charts (LSC) is a visual, executable, language for the modeling of reactive systems. Each chart depicts an inter-object scenario arising in the modeled system, partitioned into two: a monitored prechart, and a main chart. Despite the intuitive use of the language, complications arise when one wants to implement an LSC specification with decentralized components. In this paper, we introduce a variant of LSC, called distributed LSC (dLSC), which is targeted for the modeling and synthesis of decentralized systems, composed of several interacting components. While LSCs are commonly interpreted in terms of an interleaved execution of the scenarios in a sequential run, dLSCs employ partially ordered runs. We investigate the expressive power of dLSC compared to an established model of concurrent systems, namely, Petri nets, and show that dLSCs are, computationally, strictly more expressive than low-level Petri nets and subsumed by higher-level Petri nets. Specifically, we present an algorithm that synthesizes, given a dLSC specification, an equivalent token history net, which can serve as an executable implementation of the specification. Most importantly, the implementation is decentralized - components can be automatically extracted from the net. The synthesis of Petri-net components from a dLSC specification is supported by a tool.

The paradigm of Service-Oriented Computing (SOC) provides a framework for interorganizational business processes and for the emerging programming-in-the-large. The basic idea of SOC, the interaction of services, rises a lot of issues such as proper termination of interacting services or substitution of a service by another one. Such issues can be addressed by means of models of services. We show how services can intelligibly be modeled, and we present algorithms and tools to analyze properties of service models. In order to emphasize that our models properly reflect real world issues of services, we also show that services represented in established languages such as WS-BPEL can be transformed into our formal method.