User experience (UX) evaluation is widely used to assess interactive systems and support design decisions in information system engineering. Existing approaches describe UX as an effect emerging from user-product interaction and operationalize it through standardized measurement instruments. However, empirical observations suggest that UX results do not solely reflect product-related qualities but depend on evaluation prerequisites such as user needs and contextual conditions not explicitly considered in current approaches, indicating persistent interpretability limitations across diverse usage situations. This doctoral research addresses the limited scope of current UX evaluation by developing an extension that explicitly incorporates user- and context-related evaluation prerequisites. It investigates how these prerequisites shape the assessment of UX factors, what UX results represent under varying circumstances, and how structural characteristics of evaluation approaches influence result interpretability. Methodologically, the research follows the Design Science Research Methodology, grounding iterative scope development in three empirical study types: exploratory studies identify evaluation prerequisites from the user perspective, structural studies examine their relationship to existing UX measurement approaches, and explanatory studies analyze their impact on UX evaluation results in practice. The expected contribution is an empirically grounded extension of UX evaluation scope that explicates the conditions under which results can be meaningfully interpreted. By making evaluation prerequisites explicit, this research strengthens the methodological foundations of UX assessment and supports more reliable evaluation of interactive information systems.
Business Process Management (BPM) traditionally focuses on digitally mediated processes supported by Process-Aware Information Systems (PAIS). While effective in capturing control-flow events, current approaches provide limited visibility into processes unfolding within physical environments. Geospatial data, capturing real-world phenomena, is a promising tool to bridge this gap. However, existing BPM methods offer limited support for leveraging such spatio-temporal data. This research investigates how geospatial data can be integrated into BPM to bridge the gap between digital process traces and physical execution. To achieve this overarching goal, we define three objectives: (1) integrate geospatial data into PAIS to capture events and the physical execution of processes, (2) represent geospatial information in event logs while preserving spatial and temporal characteristics, and (3) embed geospatially enriched event data into process mining techniques to enable spatially-aware discovery, conformance checking, and analysis. Following an algorithm engineering methodology, this research develops artifacts to establish structured design knowledge and define a solution space for geospatially enabled BPM of real-world processes.
Modern production systems demand timely diagnostic and prognostic insights, yet the complexity of existing process intelligence (PI) tools creates a high technical barrier even for domain experts. This paper presents FIDES, a conversational neuro-symbolic tool designed to enable access to these analysis engines via natural language. Unlike approaches uniquely based on Large Language Models (LLMs) that often hallucinate operational results, FIDES implements a sound routing architecture: it uses LLMs strictly for understanding and translating the user’s intent into machine-readable encodings, while delegating the orchestration to a domain-independent automated planner. The planner autonomously decomposes complex queries and routes them to the appropriate PI engines, ensuring rigorous results. We demonstrate the tool’s maturity and usability through a lab-scale manufacturing case study, highlighting how its web-based interface enables non-technical users to perform faithful multi-perspective analysis of production processes.
Temporal requirements concern time-related aspects of information systems, such as deadlines, durations, and minimum and maximum time spans between events. Having collected a set of such temporal requirements, we need mechanisms to verify that they are not in conflict with each other, i.e., that they are temporally correct, as a conflict, for example, could lead to missing a deadline when the information system is executed. Several approaches for detecting conflicts among temporal constraints have been developed, based on temporal constraint networks as a representation model. But to date, none of the current approaches incorporates earlier knowledge about the uncertainty of an activity’s duration. This dissertation closes this gap through three major contributions. First, we will introduce a new checking approach for agile controllability, the recently introduced notion of temporal correctness, which includes earlier available information about uncertain durations. Second, we will develop an algorithm for checking agile controllability for weak contingent activities in polynomial time. Third, we will develop a modeling tool for engineers of temporal requirements that leverages the new notion of agile controllability. Overall, this research develops an agile controllability-based approach to verify temporal correctness for a broader class of temporal constraint networks than dynamic controllability, the currently prevailing temporal correctness notion, can handle.
Modernizing legacy information systems remains challenging due to fragmented knowledge, unclear decision factors, and intuitive rather than systematic approaches. This mid-stage doctoral research addresses these challenges by integrating technical, architectural, and organizational knowledge to support evidence-based modernization decisions. The research contributes: (1) a modernization knowledge ontology synthesizing findings across modernization, EA debt, change management, and compliance; (2) a factor catalog derived through systematic literature review; (3) a decision-making process extending the Renaissance framework with multi-domain considerations; and (4) a proof-of-concept LLM-supported architecture for operationalizing accumulated modernization knowledge in practice. Preliminary results include a terminology hierarchy and synonym inventory, a factor catalog under expert validation, a process with quality gates and feedback loops, and an initial RAG-based tool architecture. This work aims to enable more transparent, quantifiable, and explainable modernization decisions.
Process mining has emerged as an important data-driven approach for analyzing and improving business processes based on event log data. Although the field has traditionally focused on technical issues, growing evidence suggests that organizational and managerial factors play a critical role in shaping process mining adoption and outcomes. Recent studies have introduced maturity models and taxonomies of process mining setups, yet it remains unclear whether recurring combinations of these characteristics form distinct archetypes and how such archetypes relate to business process performance. This dissertation addresses this gap by investigating how process mining setups are structured in organizations, which archetypes can be identified, and how these archetypes differ in terms of business process performance. The study follows an exploratory sequential mixed-methods design: a qualitative phase of multiple case studies and semi-structured interviews will identify relevant setup dimensions and derive empirically grounded archetypes, followed by a survey-based quantitative phase examining recurring configurations and their relationship with business process performance. The research is currently in its early qualitative stage, with the first case organization identified and interview preparation underway. The dissertation contributes a configurational understanding of process mining adoption and its implications for business process performance.
To realize their vision, Intelligent Business Process Management Systems (BPMS) need to be able to reason upon and provide explanations grounded in complex domain knowledge. Knowledge Graph (KG) technologies have been identified as potential key building block to achieve this, by providing means to formalize, maintain, and reason upon domain knowledge. However, research into the design of KG-based BPMS is sparse. This paper presents KARIBDIS, a prototype implementation of a KG-based BPMS. KARIBDIS allows extracting diverse process knowledge from various sources and importing and integrating it in its internal KG. Further, it enables explainable decision support during process execution, grounded in the imported knowledge. We showcase the functionalities and architecture of the prototype and demonstrate its utility using a reality-inspired example process.
Object-centric process mining (OCPM) addresses the limitations of case-centric process mining by allowing events to relate to multiple objects of different types and by capturing object-to-object relationships. However, the practical adoption of OCPM is still hindered by three challenges: (1) conceptualizing object-centric event data (OCED) in a standardized way that is semantically precise and interoperable across tools, (2) investigating how OCED can be extracted from legacy information systems according to this model, including the assumptions and trade-offs involved, and (3) identifying the requirements and principles for querying OCED in a way that respects its unflattened structure. This PhD research tackles these challenges by developing an ontology-based conceptual model for object-centric event data grounded in unified foundational ontologies, investigating how ontology-based data access can support the extraction of OCED from existing information systems, and defining the semantic requirements and design principles for querying OCED. Preliminary results include an initial UFO-based metamodel that addresses limitations of current object-centric event data representations, including the current OCED Core Model, capturing object relationships, event-to-event relations, and the temporal evolution of object attributes and relations.
Hypothesis-testing behavior drives finding quality in process mining, yet failures in hypothesis testing often remain hidden from the analyst. Object-centric process mining (OCPM) amplifies this challenge by requiring analysts to test hypotheses across object types. This dissertation designs and evaluates an AI scaffold that detects hidden analytical failures during OCPM analysis and surfaces them through reflective prompting. The scaffold interacts with PM engines through standardized protocols such as the Model Context Protocol (MCP), enabling structured observation of the analyst’s reasoning. WP 1 observes how analysts with varying PM backgrounds work with object-centric event logs, identifying failure modes and expertise-differentiated patterns. WP 2 translates findings into a scaffold that detects failures algorithmically and surfaces them through reflective prompts. WP 3 evaluates using a 2 × 2 between-subjects design (PM background × scaffold condition). Contributions include: the first empirical study of hypothesis-testing behavior in OCPM; design principles for failure-detecting scaffolds; and evidence on how expertise shapes scaffold effects.
Designing data services in Cyber-Physical Production Networks (CPPN) involves managing service composition across organisational boundaries, supporting both vertical and horizontal integration, while ensuring data sovereignty, access control, and regulatory compliance. In this paper, we present D3SD (Three-layer Data Services Designer), a web-based tool for designing data services using a three-layered service model that distinguishes between different levels of the production network. D3SD provides process designers in CPPN with a BPMN-based graphical environment for modelling atomic data services that operate on data within individual supply chain actors. It also supports their composition at the smart factory and supply chain levels by leveraging a variant of the Process-to-Services (P2S) approach to suggest groupings of atomic data services that maximise internal cohesion while minimising coupling between aggregations. To ensure privacy-by-design and data sovereignty, the tool integrates Role-Based Access Control (RBAC) and GDPR responsibilities directly into the design process. We demonstrate D3SD through a multi-actor supply chain scenario derived from an industrial use case in a research project.
Complex processes can be difficult to understand from event data. Traditional process models effectively visualize control flows, but lack temporal context for performance analysis. To address this, the Performance Spectrum was introduced to map the dynamics of all steps and cases over time. However, its manual interpretation is challenging due to the overlapping patterns and noise. In this tool demonstration paper, we present PerSpect, an interactive, web-based user interface that significantly enhances the original Performance Spectrum approach. To facilitate interactive analysis, the tool automates the targeted detection of established operational behaviors (e.g., batching and FIFO), alongside advanced filtering capabilities and a side-by-side comparison function for evaluating the same segment across different process variants.
Agile software development is widely adopted due to its emphasis on collaboration and adaptability. At the same time, agile teams continue to face persistent and interrelated challenges in practice. Existing tool support addresses particular aspects of agile development but does not provide integrated support across these concerns. This paper introduces PMDAgile, a process-model-driven tool that integrates business process models into agile project environments. PMDAgile is designed to address three targeted challenges: requirements specification and communication, planning and prioritization, and change management. An exploratory evaluation with industry practitioners indicates positive perceptions of usefulness and ease of use, as well as support for the three targeted challenges, suggesting that integrating business process models into agile workflows may help mitigate these difficulties.
Process mining enables the discovery of process models from event data. However, traditional process discovery approaches often ignore contextual information or process attributes that may strongly influence real-world process behavior. This PhD research proposes a novel contextualized process discovery approach that exploits key contextual information to improve process analysis. The main idea is to use the information to enrich or abstract the discovered process model. On the one hand, the information can be used to specialize activities, resulting in contextualized process models that make context-dependent behaviors explicit. On the other hand, they can be used to group or abstract behaviours, which details may hinder process analysis. An important challenge is to find the right level of detail/abstraction that helps to answer a specific research question, i.e., finding the right trade-off between model complexity and interpretability. To support the selection and comparison of contextualized and generalized process models, the PhD also designs complexity-aware metrics to be considered within the contextualized process mining approach, including entropy-based metrics. Additionally, the PhD aims to develop tool support for interactive, context-aware process discovery that allows navigation through contextualization via zoom in/out functionality.
In conformance checking, the real-world behavior, stored in event logs, is compared with the normative behavior represented by process models. Detecting non-conforming process executions is essential to improving the information systems that support process execution and, by extension, the business process itself. Each conformance checking approach is limited by the representational bias of the process model notation used. Recent improvements in object-centric process modeling enable a more precise representation of real-world processes. Object-centric causal nets can represent concrete cardinalities, optional object types, and precise object distributions, which cannot be modeled in object-centric Petri nets. Therefore, current conformance-checking approaches based on object-centric Petri net variants cannot detect errors in these process characteristics. To enable the detection of these deviations, we propose an object-centric conformance-checking algorithm for object-centric causal nets. We provide a publicly accessible implementation of the algorithm and present a qualitative and quantitative evaluation that shows the analysis of the runtime and shows that the algorithm can detect deviations that object-centric Petri net-based approaches cannot detect.
Forecasting is a critical decision-making process for organizations, in which expert managers typically rely on the use of computational models resulting from data-driven techniques and, not infrequently, perform judgmental adjustments to reflect contextual and tacit knowledge. These adjustments, however, are knowingly prone to cognitive biases, leading to negative impacts in the decision-making process. Even though the negative impacts of cognitive biases on goals are widely recognized in the literature, forecasting experts are mostly resistant to acknowledging their occurrence due to the difficulty of understanding and quantifying this impact. This paper proposes an ontological approach for explaining and mitigating biased decisions for forecasting and operational decisions. The proposal encompasses an extension of Anchorlogy, an existing well-founded ontology about the anchoring cognitive bias, and an ontology-driven mathematical modeling component to provide expert forecasters with an explainable analysis of the quantified impacts of their decisions. The gains obtained from the proposal are analytically demonstrated using a variation of a classical problem in optimization: the news vendor problem. The validation experiment shows promising results.
Large Language Models (LLMs) are increasingly integrated into our daily lives, both through direct interaction and downstream applications. As such, it is essential that they adhere to ethical standards comparable to those expected of humans, especially when involved in decisions that affect people. Extensive research has already been conducted to evaluate whether LLMs produce biased outputs or reinforce harmful stereotypes about specific social groups. Some of this research focuses on ensuring balanced or fair outputs and decisions across different social groups. Other studies investigate how associations between identity terms and particular concepts shift when those identity terms denote different social groups. However, identity terms are not always explicitly stated in context. Attributes such as clothing or jewelry, behavior, profession, social position, location, food preferences, hobbies, coded language, and more can imply or suggest certain identities, potentially influencing decisions that would otherwise appear fair. We refer to such indirect cues as proxies—descriptions or terms that either explicitly indicate or strongly suggest a specific identity. In our research, we provide a novel and in-depth analysis of how these proxies influence the perceived fairness of LLMs.
The effectiveness of streaming process mining (PM) largely depends on the quality of its input, process event streams (PESs), continuously emitted by interconnected IT systems. A critical quality issue in real-life PESs is the occurrence of imprecise activity labels that syntactically hide contextually distinct process steps executed, i.e., homonymous activity labels. If unresolved, this issue distorts true process reality and negatively impacts the reliability and accuracy of streaming PM analyses, particularly in volatile environments. To address this challenge, we introduce DupliMend, an adaptive framework for detecting and refining homonymous activity labels within PESs by combining multi-perspective deep representation learning with online clustering to dynamically split and merge label variants as new events arrive. A hybrid drift detector monitors distributional shifts, triggering selective fine-tuning of activity-specific sparse denoising autoencoders to maintain accuracy as the PES evolves. Our continual learning mechanism reduces catastrophic forgetting and supports model adaptation while preserving learned knowledge. Extensive experiments on synthetic and real-life event logs demonstrate the effectiveness and feasibility of DupliMend, which improves cluster accuracy and precision of discovered process models with minimal latency and achieves the best average rank amongst competitor methods.
Event log quality is critical in process mining, as low-quality data can lead to inaccurate insights. Data quality issues are commonly characterized through recurring imperfection patterns in event logs. Detecting these patterns can be costly because it often requires complex data analysis logic or the involvement of domain experts in the data analysis phase. Large Language Models (LLMs) offer a promising alternative thanks to their ability to automatically generating programs that detect such patterns. In this paper, we present an agentic AI framework for detecting imperfection patterns in event logs. The framework uses carefully designed prompts that guide an LLM to generate programs for (i) identifying the patterns that most likely affect a given event log and (ii) locating the specific events responsible for these errors. These prompts are integrated into self-improving AI agents for event log data quality assessment. We evaluate the accuracy of the generated detection programs and present a proof-of-concept implementation showing how the agents iteratively improve their programs based on execution feedback.
