Optimizing business processes is crucial for improving operational efficiency. In a business process, cases arrive that require several activities to be executed in order to complete them. In process mining, methods have shown that the prefix of a case can be leveraged to predict future process states. A prefix is the historical execution trajectory of an ongoing case. However, these prefixes are not utilized by existing business process optimization approaches, limiting the available state information and thereby reducing the effectiveness of the learned resource allocation policies. This paper addresses this gap by combining the predictive power of prefixes with the optimization capabilities of Deep Reinforcement Learning (DRL) to learn resource allocation policies that minimize the mean cycle time of cases. We evaluate our approach on four synthetic processes based on real-world settings where prefix information is relevant for decision-making. The results demonstrate that including prefix information in the state space enables the DRL agent to learn policies that outperform a DRL agent without prefix information by an average of 19.5%. These findings underscore the importance of incorporating case-specific information into resource allocation methods, setting the stage for future research and real-world applications of prefix-aware process optimization.

The Dynamic Task Assignment Problem (DTAP) concerns matching resources to tasks in real time while minimizing some objectives, like resource costs or task cycle time. In this work, we consider a DTAP variant where every task is a case composed of a stochastic sequence of activities. The DTAP, in this case, involves the decision of which employee to assign to which activity to process requests as quickly as possible. In recent years, Deep Reinforcement Learning (DRL) has emerged as a promising tool for tackling this DTAP variant, but most research is limited to solving small-scale, synthetic problems, neglecting the challenges posed by real-world use cases. To bridge this gap, this work proposes a DRL-based Decision Support System (DSS) for real-world scale DTAPs. To this end, we introduce a DRL agent with two novel elements: a graph structure for observations and actions that can effectively represent any DTAP and a reward function that is provably equivalent to the objective of minimizing the average cycle time of tasks. The combination of these two novelties allows the agent to learn effective and generalizable assignment policies for real-world scale DTAPs. The proposed DSS is evaluated on five DTAP instances whose parameters are extracted from real-world logs through process mining. The experimental evaluation shows how the proposed DRL agent matches or outperforms the best baseline in all DTAP instances and generalizes on different time horizons and across instances.

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.

Predictive Process Monitoring leverages machine learning models to predict the future states of business processes. Most Predictive Process Monitoring approaches rely on black-box models which, while powerful, lack interpretability, limiting their applicability in critical decision-making scenarios. Explainable Predictive Process Monitoring has emerged to address this gap, focusing on delivering actionable and transparent insights into predictions. Current approaches, however, often fail to incorporate multiple process perspectives and granularity levels in their explanations, overlooking crucial factors that influence process outcomes. This paper proposes a novel Explainable Predictive Process Monitoring approach to deliver explanations, integrating multiple process perspectives at various levels of granularity. The proposed approach addresses the limitations of existing methods, providing comprehensive, process context-aware explanations. The effectiveness of the proposed method is assessed through experimental evaluations on real-life event logs, showing how the integration of diverse process perspectives improves the interpretability and predictive insights of local explanations with improvements ranging from 2%–3% to 15%–20% depending upon the event log under analysis.

Efficient allocation of resources to activities is pivotal in executing business processes but remains challenging. While resource allocation methodologies are well-established in domains like manufacturing, their application within business process management remains limited. Existing methods often do not scale well to large processes with numerous activities or optimize across multiple cases. This paper aims to address this gap by proposing two learning-based methods for resource allocation in business processes to minimize the average cycle time of cases. The first method leverages Deep Reinforcement Learning (DRL) to learn policies by allocating resources to activities. The second method is a score-based value function approximation approach, which learns the weights of a set of curated features to prioritize resource assignments. We evaluated the proposed approaches on six distinct business processes with archetypal process flows, referred to as scenarios, and three realistically sized business processes, referred to as composite business processes, which are a combination of the scenarios. We benchmarked our methods against traditional heuristics and existing resource allocation methods. The results show that our methods learn adaptive resource allocation policies that outperform or are competitive with the benchmarks in five out of six scenarios. The DRL approach outperforms all benchmarks in all three composite business processes and finds a policy that is, on average, 12.7% better than the best-performing benchmark.

Business process event logs consist of the actual process followed by employees of a business by which the event log was generated. Activities in a business process are completed in a non-uniform manner and in settings where fault occurrences are inevitable. Such faults, or deviation from the defined process, can be considered anomalies in the process data. Over the years, several anomaly detection methods have been proposed to detect and filter out such anomalies from an event log. However, little attention has been given to finding out what caused these anomalies and why they were considered anomalies. This paper focuses on this unaddressed topic and proposes a method to provide model-agnostic post-hoc explanations of the detected anomalies by using linguistic summaries.

Outcome-oriented predictive process monitoring aims at classifying a running process execution according to a given set of categorical outcomes, leveraging data on past process executions. Most previous studies employ Recurrent Neural Networks to encode the sequence of events, without taking the structure of the process into account. However, process executions typically involve complex control-flow constructs, like parallelism and loops. Different executions of these constructs can be recorded as different event sequences in the event log. This makes it challenging for a recurrent classifier to detect potential relations between a high-level control-flow construct and the prediction target. This is especially true in the presence of high variability in process executions and lack of data. In this paper, we propose a novel approach which encodes the control-flow construct each event belongs to. First, we exploit Local Process Model mining techniques to extract frequently occurring control-flow patterns from the event log. Then, we employ different encoding techniques to enrich an on-going process execution with information related to the extracted control-flow patterns. We tested the proposed method on nine real-life event logs. The obtained results show consistent improvements in the prediction performance.

Quantitative business process analysis is a powerful approach for analyzing timing properties of a business process, such as the expected waiting time of customers or the utilization rate of resources. Multiple techniques are available for quantitative business process analysis, which all have their own advantages and disadvantages. This paper presents a novel technique, based on queueing models, that combines the advantages of existing techniques, in that it leads to accurate analyses, is computationally inexpensive, and feature complete with respect to its support for basic process modeling constructs. An extensive quantitative evaluation has been performed that compares the presented queueing model to existing queueing models from literature. This evaluation shows that the presented model outperforms existing models with one order of magnitude on accuracy. The resulting queueing model can be used for fast and accurate timing predictions of business process models. These properties are useful in optimization scenarios.

Traditionally, the optimal resource allocation in a business process is determined by manually exploring a number of options, but whether this leads to the optimal solution is questionable; it is possible that an unexplored option produces even better results. Therefore, an automated approach is needed that explores all possible options in a structured manner to find the best one. To address this need this paper proposes three iterations of a search strategy for resource optimization that searches through the space of possible resource allocations to find the best one. The performance of the search strategy is evaluated theoretically with guided experiments. The experiments show that the proposed search strategies scale very well and obtain the optimal value in only a fraction of the number of steps needed by an exhaustive exploration of the solution space. In this manner, this paper also forms the basis for future research into resource optimization with extensions into multi-process optimization, advanced, and operational resource optimization.

In this paper we propose a Deep Reinforcement Learning approach to solve a multimodal transportation planning problem, in which containers must be assigned to a truck or to trains that will transport them to their destination. While traditional planning methods work "offline" (i.e., they take decisions for a batch of containers before the transportation starts), the proposed approach is "online", in that it can take decisions for individual containers, while transportation is being executed. Planning transportation online helps to effectively respond to unforeseen events that may affect the original transportation plan, thus supporting companies in lowering transportation costs. We implemented different container selection heuristics within the proposed Deep Reinforcement Learning algorithm and we evaluated its performance for each heuristic using data that simulate a realistic scenario, designed on the basis of a real case study at a logistics company. The experimental results revealed that the proposed method was able to learn effective patterns of container assignment. It out-performed tested competitors in terms of total transportation costs and utilization of train capacity by 20.48% to 55.32% for the cost and by 7.51% to 20.54% for the capacity. Furthermore, it obtained results within 2.7% for the cost and 0.72% for the capacity of the optimal solution generated by an Integer Linear Programming solver in an offline setting.

Prognostics is the area of research that is concerned with predicting the remaining useful life of machines and machine parts. The remaining useful life is the time during which a machine or part can be used, before it must be replaced or repaired. To create accurate predictions, predictive techniques must take external data into account on the operating conditions of the part and events that occurred during its lifetime. However, such data is often not available. Similarity-based techniques can help in such cases. They are based on the hypothesis that if a curve developed similarly to other curves up to a point, it will probably continue to do so. This paper presents a novel technique for similarity-based remaining useful life prediction. In particular, it combines Bayesian updating with priors that are based on similarity estimation. The paper shows that this technique outperforms other techniques on long-term predictions by a large margin, although other techniques still perform better on short-term predictions.

Process mining can be used to analyze business processes based on logs of their execution. These execution logs are often obtained by querying a database and storing the results in a file. The mining itself is then done on the file, such that the data processing power of the database cannot be used after the log is extracted. Enabling process mining directly on a database therefore provides additional flexibility and efficiency. To help facilitate this, this paper formally defines a database operator that extracts the ‘directly follows’ relation—one of the relations that is at the heart of process mining—from an operational database. It defines the operator using the well-known relational algebra and formally proves equivalence properties of the operator that are useful for query optimization. Subsequently, it presents time-complexity properties of the operator. Finally, it presents an implementation of the operator as part of the H2 DBMS and demonstrates that this implementation extracts the ‘directly follows’ relation from a database with an arbitrary database structure within a fraction of a second; several orders of magnitude faster than is currently possible.

A growing number of business process management systems is under development both in academia and in practice. These systems typically are based on modern system engineering principles, such as service-oriented architecture. At the same time, the advent of big data analytics has changed the scope of these systems, including functionality such as data mining. However, existing reference architectures for business process management systems date back 20 years and, consequently, are not up-to-date with these modern developments. To fill the gap, this article proposes an up-to-date reference architecture, called BPMS-RA, for modern business process management systems. BPMS-RA is based on analysis of recent literature and of existing commercial implementations. This reference architecture aims to provide a guideline template for the development of modern-day business process management systems by specifying functions and interfaces that need to be provided by these systems as well as a set of quality criteria that they need to meet.

Simulation is often used to provide quantitative predictions on the performance of business processes. However, the current available business process simulation engines only support basic resource constructs and even these are sometimes lacking. This leads to simulated performance metrics that can deviate significantly from the real process performance. This paper provides an overview of the quantitative effects of incorporating these advanced resource constructs in business process simulation. Experiments are conducted to assess whether there are significant differences between the effects of using basic resource constructs and using their advanced counterparts. The main conclusions are: that the allocation strategy deployed by the resources is of significant influence on the performance of the process; that the ability to assign resources to multiple roles has a significant influence; and that simulation replications are absolutely necessary to produce reliable simulation results. These insights can be used by researchers and practitioners to understand the validity of the results of their simulation studies and as a direction for future research.

Quantitative analysis of business processes facilitates the computation of quantitative properties of a business process, such as the average sojourn time and cost of cases of the process. Moreover, it allows for what-if analysis, by enabling the computation of such properties for alternative business process designs, without requiring those designs to be implemented in practice. This paper presents a novel technique for quantitative analysis that is both fast and generally applicable. The technique is developed as an extension of the state-of-the-art in queueing network analysis, by lifting some of the restrictions that are imposed by existing techniques. In particular, it enables precise quantitative analysis of business process models that contain many-to-many relations between tasks and resources (via roles). As a consequence, the technique presented in this paper allows for quantitative analysis that is more generally applicable than existing queueing network analysis techniques and only takes a fraction of the execution time required by existing simulation techniques. In doing so, it paves the way to advanced analysis techniques, such as prescriptive analytics.

Generation of linguistic summaries that are compact, short and relevant to the user remains an open challenge. In this paper, we propose a novel method for improving the generation of linguistic summaries inspired by the a-priori algorithm and the degree of appropriateness. The method generates all true summaries with related predicates in the summarizer, resulting in a small set of linguistic summaries, whose presentation to the user is compact. We tested our method on three real world data sets. The results indicate that our proposed approach is a good alternative to previous methods suggested for generating linguistic summaries.

Process discovery algorithms aim to capture process models from event logs. These algorithms have been designed for logs in which the events that belong to the same case are related to each other — and to that case — by means of a unique case identifier. However, in service-oriented systems, these case identifiers are rarely stored beyond request-response pairs, which makes it hard to relate events that belong to the same case. This is known as the correlation challenge. This paper addresses the correlation challenge by introducing a technique, called the correlation miner, that facilitates discovery of business process models when events are not associated with a case identifier. It extends previous work on the correlation miner, by not only enabling the discovery of the process model, but also detecting which events belong to the same case. Experiments performed on both synthetic and real-world event logs show the applicability of the correlation miner. The resulting technique enables us to observe a service-oriented system and determine — with high accuracy — which request-response pairs sent by different communicating parties are related to each other.

Linguistic summarization techniques make it easy to gain insight into large amounts of data by describing the main properties of the data linguistically. In this paper we focus on a specific type of data, namely process data, i.e., event logs that contain information about when some activities were performed for a particular customer case. An event log may contain many different sequences, because actions or events are often performed in slightly different orders for different customer cases. This easily leads to a very large number of generated summaries. As the point of linguistic summarization is to provide a quick overview, such a large number of summaries is not helpful to the interpreter. To address this problem, we propose a method for the generation of linguistic summaries of sequences that groups similar sequences and returns them in a single linguistic summary. We show the applicability of our technique on an event log from practice and show that it can be used to reduce the number of produced summaries by 80%, while keeping the important information that is contained in those summaries.

In this paper we investigate a new application area for protoform-based linguistic summaries, namely process data. We are proposing four protoforms that are designed to capture process specific information. Linguistic summaries can capture information on the tasks or sequences of tasks that are frequently executed as well as properties of these tasks or sequences, such as their throughput and service time. Such information is of specific interest in the context of process analysis and diagnosis. Moreover linguistic summaries are capable of combining different features and characteristics of tasks, sequences of tasks and even whole processes. Through a case study with a data from practice, the paper shows that the knowledge derived from these linguistic summaries is useful for identifying problems in processes and establishing best practices.

During the execution of a service collaboration, a party may drop out for technical reasons or business reasons. In that case, that party must be replaced in the collaboration, at run-time, by a new party. Ideally, the new party can pick up where the old party left. Currently, algorithms exist that can help with the selection and adaptation of the new party to incorporate it in the collaboration. Also, algorithms exist that can help to pick up a business process where it left off. However, to the best of our knowledge, no algorithms exist that can help a new party in a collaboration to pick up where the old party left off. This paper fills that gap, by providing an overview of the components and operations that are necessary to enable a party in a collaboration to be replaced by another party at run-time. In addition the paper presents two strategies, and the corresponding algorithms, that realize the architecture. As a proof-of-concept, a tool was developed that implements both strategies.

This article addresses the problem of constructing consolidated business process models out of collections of process models that share common fragments. The article considers the construction of unions of multiple models (called merged models ) as well as intersections (called digests ). Merged models are intended for analysts who wish to create a model that subsumes a collection of process models -- typically representing variants of the same underlying process -- with the aim of replacing the variants with the merged model. Digests, on the other hand, are intended for analysts who wish to identify the most recurring fragments across a collection of process models, so that they can focus their efforts on optimizing these fragments. The article presents an algorithm for computing merged models and an algorithm for extracting digests from a merged model. The merging and digest extraction algorithms have been implemented and tested against collections of process models taken from multiple application domains. The tests show that the merging algorithm produces compact models and scales up to process models containing hundreds of nodes. Furthermore, a case study conducted in a large insurance company has demonstrated the usefulness of the merging and digest extraction operators in a practical setting.

Once multiple models of a business process are created for different purposes or to capture different variants, verification of behaviour equivalence or compatibility is needed. Equivalence verification ensures that two business process models specify the same behaviour. Since different process models are likely to differ with respect to their assumed level of abstraction and the actions that they take into account, equivalence notions have to cope with correspondences between sets of actions and actions that exist in one process but not in the other. In this paper, we present notions of equivalence and compatibility that can handle these problems. In essence, we present a notion of equivalence that works on correspondences between sets of actions rather than single actions. We then integrate our equivalence notion with work on behaviour inheritance that copes with actions that exist in one process but not in the other, leading to notions of behaviour compatibility. Compatibility notions verify that two models have the same behaviour with respect to the actions that they have in common. As such, our contribution is a collection of behaviour equivalence and compatibility notions that are applicable in more general settings than existing ones.

Nowadays, it is common for organizations to maintain collections of hundreds or even thousands of business processes. Techniques exist to search through such a collection, for business process models that are similar to a given query model. However, those techniques compare the query model to each model in the collection in terms of graph structure, which is inefficient and computationally complex. This paper presents an efficient algorithm for similarity search. The algorithm works by efficiently estimating model similarity, based on small characteristic model fragments, called features. The contribution of this paper is threefold. First, it presents three techniques to improve the efficiency of the currently fastest similarity search algorithm. Second, it presents a software architecture and prototype for a similarity search engine. Third, it presents an advanced evaluation of the algorithm. Experiments show that the algorithm in this paper helps to perform similarity search about 10 times faster than the original algorithm.

This paper studies the following problem: given a pair of business process models, determine which elements in one model are related to which elements in the other model. This problem arises in the context of merging different versions or variants of a business process model or when comparing business process models in order to display their similarities and differences. The paper investigates two approaches to this alignment problem: one based purely on lexical matching of pairs of elements and another based on error-correcting graph matching. Using a set of models taken from real-life scenarios, the paper empirically shows that graph matching techniques yield a significantly higher precision than pure lexical matching, while achieving comparable recall.

To unify similar business processes, such as processes of similar business units or similar organizations, the similarities and differences between these business processes must be detected and the differences must be resolved. This paper presents a classification that describes frequently occurring differences between business processes. This classification is helpful in the difference detection step. We discovered the differences in practice by comparing processes that we obtained from different business units in two organizations. The classification helps when merging processes in case of a merger between organizations. It also helps when merging processes to construct a standardized process and when verifying whether a party adheres to that standardized process.

In this paper we analyse the interaction mechanisms provided by Web services technology and by CORBA. Specifically we analyse the request/response, callback, polling and (multicast) message passing mechanisms. As a result we present coloured Petri nets that capture the behaviour of these mechanisms precisely. Based on our analysis we define concepts for representing the Web services and CORBA interactions in a suitable and platform independent manner. These concepts can be used for platform independent design of distributed applications, while they (provably) maintain the consistency with platform specific implementations. Because their behaviour is defined by Petri nets, the concepts also support simulation, validation and verification of designs. We also evaluate the suitability of UML's concepts for representing the mechanisms and the degree of platform independence that these concepts can achieve

In a model-driven design process the interaction between application parts can be described at various levels of platform-independence. At the lowest level of platform-independence, interaction is realized by interaction mechanisms provided by specific middleware platforms. At higher levels of platform-independence, interaction must be described in such a way that it can be further refined and realized onto a number of different middleware platforms, each with its particular interaction mechanisms and implementation constraints. In this paper, we investigate concepts that support interaction design at various levels of middleware-platform-independence. In addition, we propose design operations for interaction refinement. The application of these operations to source designs results in target designs that take into account implementation constraints imposed by platforms, while preserving characteristics prescribed in source designs. Target designs are related to source designs by conformance. We discuss how transformation and conformance can be related, such that transformations indeed preserve the characteristics prescribed by a source design.

In a model-driven design process the interaction between application parts can be described at various levels of platform-independence. At the lowest level of platform-independence, interaction is realized by interaction mechanisms provided by specific middleware platforms. At higher levels of platform-independence, interaction must be described in such a way that it can be further refined and realized onto a number of different middleware platforms, each with its particular interaction mechanisms and implementation constraints. In this paper, we investigate concepts that support interaction design at various levels of middleware-platform-independence. Also, we propose design operations for interaction refinement. The application of these operations to source designs results in target designs that take into account implementation constraints imposed by platforms, while preserving characteristics prescribed in source designs.

This paper presents a service-oriented design approach that allows one to relate services modelled at different levels of granularity during a design process, such as business and application services. To relate these service models we claim that a 'concept gap' and an 'abstraction gap' need to be bridged. The concept gap represents the difference between the conceptual models used to construct service models by different stakeholders involved in the design process. The abstraction gap represents the difference in abstraction level at which service models are defined. Two techniques are presented that bridge these gaps. Both techniques are based on the Interaction System Design Language (ISDL). The paper illustrates the use of both techniques through an example.

Currently, service-oriented computing is mainly technology-driven. Most developments focus on the technology that enables enterprises to describe, publish and compose application services, and to communicate with applications of other enterprises according to their service descriptions. In this paper, we argue that this technology should be complented with modelling languages, design methods and techniques supporting service-oriented design. We consider service-oriented design as the process of designing application support for business processes, using the service-oriented paradigm. We assume that service-oriented computing technology is used to implent application support. The paper presents two main contributions to the area of service-oriented design. First, a systatic service-oriented design approach is presented, identifying generic design milestones and a method for assessing the conformance between application designs at related abstraction levels. Second, a conceptual model for service-oriented design is presented that provides a common and precise understanding of the terminology used in service-oriented design. The ISDL modelling language is introduced to express service-oriented designs, based on this conceptual model. The paper includes an elaborate example to illustrate our ideas.

As the technology associated with the "Web Services" trend gains significant adoption, the need for a corresponding design approach becomes increasingly important. This paper introduces a foundational model for designing (composite) services. The innovation of this model lies in the identification of four interrelated viewpoints (interface behaviour, provider behaviour, choreography, and orchestration) and their formalization from a control-flow perspective in terms of Petri nets. By formally capturing the interrelationships between these viewpoints, the proposal enables the static verification of the consistency of composite services designed in a cooperative and incremental manner. A proof-of-concept simulation and verification tool has been developed to test the possibilities of the proposed model.

The architectural design of distributed enterprise applications from the viewpoints of different stakeholders has been proposed for some time, for example, as part of RM-ODP and IEEE 1471, and seems now-a-days to gain acceptance in practice. However, much work remains to be done on the relationships between different viewpoints. Failing to relate viewpoints may lead to a collection of viewpoint models that is inconsistent, and may therefore lead to an incorrect implementation. This paper defines an approach that helps designers to relate different viewpoints to each other. Thereby, it helps to enforce the consistency of the overall design. The results of this paper are expected to be particularly interesting for Model Driven Architecture (MDA) projects, since the proposed models can be used for the explicit definition of the models and relationships between models in an MDA trajectory.