The 10 most recently published documents
Marketing text analysis often relies on narrow, specialized datasets: too limited for generic large-scale NLP, yet too large for manual review. This dissertation introduces labeled UD-LDA, a topic model that conditions topic propagation on Universal Dependencies, allowing topics to flow along syntactic relations rather than relying solely on word order or co-occurrence. Applied to three customer review datasets, it outperforms standard benchmarks, yielding better model fit and more distinct yet coherent topics. Modeling propagation as a function of dependency type reveals that modifiers and function words promote topic consistency, while relations linking distinct syntactic units suppress it, showing that grammar systematically structures latent thematic content.
Is what I see what I become? observational leadership development in management consultancies
(2026)
Leadership development is often approached as an individual process focused on traits, competencies, and formal learning interventions. This dissertation reconceptualizes leadership development as a socially embedded, observational, and contextually shaped process, termed Observational Leadership Development.
With particular attention to management consultancies as dynamic professional environments characterized by frequent exposure to diverse leader role models, the dissertation draws on social learning theory, identity theory, and qualitative analysis to examine how individuals learn leadership through observing, interpreting, and selectively integrating the behaviors, values, and practices of multiple leader role models.
Across three studies, it develops a conceptual understanding of leader role models, identifies key characteristics, and shows how career ambitions, organizational context, and career tenure influence role model selection and leadership learning. It further identifies three reflective processing modes — situational adaptation, selective integration, and contrastive delineation — that explain how observed leadership experiences are translated into leader identity formation. Building on these insights, the dissertation proposes a Leadership Learning Cycle that captures the recursive interplay of observation, reflection, identity work, and contextual influence in leadership development.
Die Gestaltung logistischer Netzwerke stellt für Handelsunternehmen eine zentrale strategische Aufgabe dar. Breite und heterogene Sortimente, räumlich verteilte Lieferantenstrukturen, dichte Netze nachgelagerter Nachfragepunkte sowie die zunehmende Verzahnung stationärer und digitaler Vertriebskanäle führen zu ausgeprägten Interdependenzen zwischen Beschaffung, Lagerinfrastruktur und Distribution. Die vorliegende kumulative Dissertation leistet einen Beitrag zu einer integrierten Konzeptualisierung und Modellierung des strategischen Supply Chain Network Design in der Handelslogistik. Sie umfasst vier Einzelbeiträge.
Der erste Beitrag entwickelt einen mehrschichtigen konzeptionellen Bezugsrahmen, der Kontextbedingungen, strategische Gestaltungsentscheidungen, Netzwerkkonfigurationen und Leistungswirkungen handelslogistischer Netzwerke systematisiert und Resilienz, Nachhaltigkeit, Kollaboration sowie Omni-Channel-Retailing als Querschnittsdimensionen integriert. Die drei weiteren Beiträge entwickeln gemischt-ganzzahlige Optimierungsmodelle für zentrale Strukturentscheidungen: die integrierte Wahl von Distributionszentrumstypen und Produktzuordnung, die simultane Gestaltung der Breite und Tiefe von Distributionsnetzwerken (funktionale Differenzierung und Zentralisierungsgrad) sowie die integrierte Planung von Cross-Dock-Standorten, Belieferungsmodi und Liefermustern auf der Beschaffungsseite. Aufgrund der Komplexität der Probleme werden hierarchische Dekompositionsverfahren und heuristische Lösungsansätze vorgeschlagen. Fallstudien mit Realdaten eines großen europäischen Einzelhandelsunternehmens belegen die praktische Anwendbarkeit und quantifizieren erhebliche Effizienzpotenziale.
Die Dissertation zeigt insgesamt, dass strategisches Netzwerkdesign im Handel als integriertes, mehrdimensionales Strukturproblem zu verstehen ist, dessen Teilentscheidungen nur im Zusammenspiel adäquat bewertet werden können.
Überlegungen zur menschlichen Natur werden bereits seit Jahrtausenden diskutiert. Ausgehend von philosophischen Gedanken aus der Antike werden Fragen nach der Natur des Menschen über die Philosophie hinausgehend in vielen unterschiedlichen Wissenschaften aufgegriffen, wie die Theologie, Biologie, Soziologie, Politikwissenschaft, Wirtschaftswissenschaften, Anthropologie und auch die Psychologie. Wahrend es in der Philosophie um normativ-philosophische Annahmen geht, stehen innerhalb der Psychologie subjektive Annahmen im Vordergrund, die mittels empirischer Methoden untersucht und erhoben werden. Auch auf dieser Grundlage wurden, genau wie in den anderen genannten Disziplinen auch, unterschiedliche Menschenbildannahmen proklamiert, die die implizite Grundlage ebenso unterschiedlicher wissenschaftlicher Strömungen, Theorien und Modelle bilden.
Menschenbilder werden in dieser Arbeit aus psychologischer Sicht betrachtet und untersucht. Damit wird die subjektive Seite in den Vordergrund gestellt wird. Denn überraschenderweise gibt es - auch innerhalb der Psychologie - weitaus weniger empirische Forschung zu Menschenbildern und Menschenbildannahmen, als man angesichts der Bedeutung des Forschungsfeldes erwarten könnte. In dieser Arbeit ist von Menschenbildannahmen die Rede, die die Subjektivität der Annahmen betonen, die sich zugleich empirisch betrachten und erheben lassen und die die Vereinbarkeit vielfältiger Annahmen über die menschliche Natur ermöglichen, statt sich auf ein spezifisches und fest definiertes Menschenbild festzulegen.
Menschenbildannahmen sind definiert als Annahmen von Individuen über Menschen im Allgemeinen und umfassen die subjektive Zuschreibung vielfältiger Attribute wie Eigenschaften, Fähigkeiten und Neigungen zur menschlichen Natur. Damit sind diese Attribute subjektiv grundsätzlich auf alle Menschen anwendbar. Es ist denkbar, dass das Maß, in welchem diese Annahmen von einem Individuum auf bestimmte Personen angewandt werden, davon abhängt, wie viele spezifische Informationen dem Individuum zu der betreffenden Person zur Verfügung stehen.
In Studie 1 (N = 333) wurde, aufbauend auf den Attributen, die im Diskurs zu vermeintlich allgemeingültigen Menschenbildern der menschlichen Natur zugeschrieben wurden, ein Skaleninventar entwickelt, mit dem subjektive Menschenbildannahmen in Fragebogenstudien gemessen werden können. Dieses Skaleninventar wurde in Studie 2 (N = 525) zunächst auf Basis positiver Korrelationen mit verwandten Konstrukten konvergent validiert und anschließend ihr Zusammenhang mit kooperativen Konfliktlösungsverhalten untersucht. Die Ergebnisse darauf hin, dass die Annahme, dass Menschen von Natur aus beziehungsorientiert sind, vermittelt über Empathie, Verantwortung, Beziehungszufriedenheit und Kontakthäufigkeit, einen positiven Zusammenhang mit kooperativem Konfliktlösungsverhalten haben kann. Auch die Annahmen, dass Menschen inhärent eigennützig und hilfsbereit sind, tragen zur Erklärung dieser Konflikt-variable bei. Hierauf wurde in Studie 3 (N = 906) ein erster empirischer Hinweis gefunden, in der kooperatives Konfliktlösungsverhalten mittels der Theorie des geplanten Verhaltens erklärt wurde. Hier haben die beiden Annahmen jeweils einen moderierenden Effekt. Zuletzt wurde in Studie 4 (N = 649) längsschnittlich untersucht, inwiefern Menschenbildannahmen einen Einfluss auf die Präferenzen für Faschismus, Konservatismus, Liberalismus, Sozialismus und Ökologismus haben: Annahmen über die menschliche Natur tragen hier zur Erklärung dieser fünf politischen Einstellungen bei, wobei die Effekte von soziodemographischen Variablen moderiert werden.
In der Gesamtschau zeigt die Studienreihe, dass der psychologische Ansatz der Menschenbildannahmen Subjektivität und Vielfältigkeit der Annahmen über die menschliche Natur vereint. Er ermöglicht damit die gemeinsame empirische Erforschung dieser Annahmen. Die ersten empirischen Befunde aus dieser Arbeit sprechen für die Relevanz von Menschenbildannahmen in psychologischen Erklärungsmodellen. Damit legt diese Arbeit einen Grundstein, um mit der empirischen Forschung zu subjektiven Menschenbildannahmen psychologische Erklärungsmodelle in verschiedenen Kontexten zu ergänzen und die Forschung zur menschlichen Natur um eine neue Perspektive zu erweitern.
This dissertation investigates fairness and efficiency challenges in urban logistics systems. The research combines mathematical optimization models, heuristic solution approaches, and machine learning methods to support decision-making in collaborative logistics networks and cost allocation problems.
A first focus of the dissertation is the collaboration of logistics service providers in two-tier city logistics systems. To this end, mathematical models are developed that integrate strategic and operational planning decisions while accounting for both economic and environmental objectives. Since the resulting optimization problems cannot be solved exactly for realistically sized instances, specialized metaheuristic solution methods are proposed to generate high-quality solutions within reasonable computation times. The results demonstrate that collaboration can substantially reduce transportation costs and emissions, while simultaneously raising important questions regarding the fair distribution of benefits and responsibilities among participating stakeholders.
Building upon this foundation, the dissertation examines different fairness concepts in collaborative logistics systems. Fairness requirements related to cost allocation, workload distribution, and service regularity are incorporated into mathematical planning models. The computational experiments show that fairness constraints significantly influence system performance and that evaluating fairness over longer planning horizons can provide a balanced trade-off between economic, environmental, and social objectives.
A second major focus of the dissertation concerns the allocation of routing costs using the Shapley value from cooperative game theory. While the Shapley value is widely regarded as one of the fairest allocation mechanisms, its exact computation becomes computationally prohibitive for large-scale optimization problems. To address this challenge, two approximation approaches are developed. The first employs machine learning techniques to predict Shapley values based on structural characteristics of optimization problems. The second is a problem-independent approach that approximates Shapley values through the aggregation of exact values computed for smaller subcoalitions. Computational results demonstrate that both approaches achieve high approximation accuracy while significantly reducing computation times.
Overall, the dissertation contributes to the development of fair and efficient decision-support methods for urban logistics systems. By combining techniques from operations research, cooperative game theory, and machine learning, the proposed models and algorithms provide both theoretical insights and practical guidance for the design of sustainable and collaborative logistics systems.
The present cumulative thesis is concerned with theoretical properties of (deep) fully connected neural networks. It is based on five publications and can be divided into two parts.
In the first part, we establish bounds on the ℓᵖ-Lipschitz constants of deep ReLU neural networks at initialization, that is, networks with random weights and biases. More precisely, we assume that the weights and biases are drawn according to a generalization of the popular He initialization. In the zero-bias case, we prove upper and lower high-probability bounds for wide networks that differ only by a logarithmic factor in the network width and a polynomial factor in the network depth. We then extend the analysis to symmetric bias distributions, for which similar bounds are established.
The second part of the thesis investigates the approximation capabilities of complex-valued neural networks (CVNNs), that is, networks with complex-valued weights and biases and with activation functions mapping from ℂ to ℂ. We prove sharp quantitative bounds for the worst-case approximation error when approximating Cʳ-functions by shallow CVNNs, both under continuous weight selection and without any assumptions on the weight selection. As part of this analysis, we generalize well-known results on the approximation properties of univariate ridge functions to the multivariate setting. Moreover, we study the universality of deep, narrow CVNNs, that is, classes of CVNNs with restricted width but arbitrary depth.
This dissertation examines the role of tax policy in enhancing a country’s attractiveness as a business location and promoting corporate investment, innovation, and economic growth. Against the backdrop of declining economic growth in Germany and increasing structural challenges, the study investigates whether tax incentives and efficient tax frameworks can stimulate private investment and strengthen international competitiveness.
The dissertation consists of three empirical studies. Chapter 1 analyzes the effects of tax incentives for automation using a quasi-natural experiment based on a reduction in South Koreas automation tax credit and a difference-in-differences approach combining robot and firm-level data.
Chapter 2 evaluates the effectiveness of indirect tax incentives for research and development (R&D), focusing on Germany’s research allowance and drawing on both existing empirical evidence and innovation data.
Chapter 3 examines the economic consequences of tax complexity using an international sample of publicly listed firms and a two-way fixed effects regression framework.
The findings show that tax incentives significantly influence corporate investment decisions. Automation tax incentives increase investment in automation but may also encourage inefficient over investment, while their reduction leads to lower automation investment and higher employment. R&D tax incentives stimulate research expenditure, innovation output, productivity, and R&D-related employment. At the same time, the results demonstrate that tax complexity imposes substantial economic costs by reducing firm value, increasing compliance burdens, and weakening the effectiveness of tax incentives.
Overall, the dissertation highlights that tax policy can contribute to economic growth and location attractiveness when it simultaneously provides targeted investment incentives and maintains a transparent and efficient tax system. The findings offer important implications for the design of tax policy aimed at fostering investment, innovation, and long-term competitiveness.
Four essays on the future of robotic process mining and its influence on robotic process automation
(2026)
Despite significant technological advancements, the development lifecycle of Robotic Process Automation (RPA) bots remains heavily labor-intensive and difficult to scale. Robotic Process Mining (RPM) has emerged to automate this lifecycle by discovering routine behaviors directly from user interaction (UI) logs. However, current state-of-the-art RPM tools depend on clean, pre-segmented recordings, effectively shifting the burden of task isolation onto the user. In reality, unsegmented real-world UI logs are highly unstructured, contextually diverse, and filled with intra- and inter-routine noise.
This cumulative dissertation addresses these limitations through a strategic and operational enhancement of the RPM pipeline across four research essays. First, it examines how human developer decision-making shapes RPA bot programming and maps these practices to identify gaps in current automation tools. To bridge the semantic gap between recorded actions and deployable code, an ontology-driven conceptual framework is introduced to formalize user actions and RPA building blocks. The technical core of the dissertation introduces two novel unsupervised time-series data mining approaches designed to decouple routine discovery from human intervention. By leveraging Word2Vec encoding, matrix profiles, and grammar induction, these methods adaptively segment variable-length automation candidates directly from continuous, high-noise work sessions. Ultimately, the artifacts presented in this work resolve critical semantic and scalability bottlenecks, paving the way for fully autonomous and resilient hyper-automation enterprise capabilities.
A bio-psycho-social perspective on anxiety disorders treated with cognitive behavioral therapy
(2026)
Anxiety disorders (ADs) are one of the most prevalent mental disorders, affecting up to around 30% of the population worldwide (Jacobi et al., 2014; Szuhany & Simon, 2022). They represent pathological anxiety and fear, marked by disproportionate emotional and physiological responses in the absence of real threat, leading to avoidance behavior or safety strategies. They often include panic attacks with cardiovascular and respiratory symptoms. ADs impair functioning, reduce quality of life, and impose a major socioeconomic burden (Baxter et al., 2014; Santomauro et al., 2021). The pathogenesis of AD is multifactorial, including biological, psychological, and environmental causes. Cognitive behavioral therapy (CBT), integrating exposure to feared situations and targeting maladaptive beliefs, is a widely used and highly effective treatment for various AD (Bandelow et al., 2023; Hofmann et al., 2025). Nevertheless, approximately one-third of patients do not respond to treatment. Identifying underlying mechanisms of action for CBT treatments and disentangling biological and psychological risk factors would allow a better understanding of the etiology of ADs. Moreover, it would allow to determine the utility of biomarkers in predicting therapy response and relapse risk as well as the development of a more personalized “precision” medicine to tailor treatments to the individual's needs, which has the potential to improve treatment response rates. Hence, recent research efforts have focused more on individual differences to identify relevant factors for the prediction of therapy response: biological mechanisms, such as (epi-)genetics and stress hormones, environmental factors, such as maltreatment experiences or resulting attachment styles, as well as therapy process variables, such as therapeutic alliance (Bandelow et al., 2016; Bandelow et al., 2017; Fava & Morton, 2009; Zilcha-Mano & Fisher, 2022). However, our understanding of how psychotherapy might be influenced by, e.g., (epi-)genetics and the biological stress response, is still limited. Even less is known about the impact of CBT on biological processes. Understanding the impact of CBT as a positive and predictable environmental factor becomes even more relevant as many studies have highlighted how intertwined the environment, e.g., maltreatment experiences, is with biological mechanisms, e.g., related to stress response (Fischer et al., 2021) and epigenetics (Weber et al., 2025). Interestingly, attachment style has also emerged as a relevant risk factor for the development of ADs (Levy et al., 2011). Further, it seems likely that attachment style influences the psychotherapy success of AD via the therapeutic alliance (for a meta-analysis see Notsu et al., 2025). Hence, it seems important to understand how CBT for AD is influenced by and may influence biological mechanisms and relational processes. An integration of research results from various domains within a bio-psycho-social model could contribute to a better understanding of the contributions of several mechanisms in therapy. Aiming to address this research gap, we conducted the following studies.
In Publication I, we conducted a literature review of how CBT might be related to and might influence the functioning of the hypothalamic-pituitary-adrenal (HPA) axis in patients with ADs. Since it is one of the most relevant biological systems for stress response, understanding changes within this system is particularly relevant for patients suffering from ADs. Overall, we found evidence supporting the idea that cortisol levels have the potential to indicate the AD disease status and serve as a possible biomarker in therapy response prediction. In detail, the reviewed studies suggest that persistently elevated cortisol levels seem to impede therapy response. We further focused on how CBT, specifically therapeutic exposure to feared situations, might influence HPA axis functioning. A beneficial effect of cortisol elevation during exposure or, in reverse, effects of a blunted HPA system response on an unfavorable outcome are supported. Moreover, effective CBT treatment seems to normalize HPA system hyperfunction. Additionally, cognitive processes, e.g., positive anticipatory appraisal, attribution, perceived controllability, and coping, seem to have an impact on adaptive HPA system functioning.
Second, we conducted a therapy progress study with patients suffering from panic disorder (PD) undergoing CBT at the outpatient clinic of the Max Planck Institute of Psychiatry in Munich. We assessed dynamic changes in epigenetic (gene methylation and expression), metabolomic (small molecules produced by metabolism), and immune system biomarkers, as well as in relational patterns (attachment style, interpersonal distress, and therapeutic alliance). We applied several measures, including various questionnaires and taking blood samples, during a standardized short-term exposure-based CBT (12 sessions and two booster sessions) at several time points: pre-/ post-, mid-treatment, exposure sessions, and follow-up.
Biological markers on different levels were analyzed: Publication II identified specific DNA methylation signatures (HECA gene), which, as time-stable heritable gene-regulatory mechanisms, might serve as biomarkers for PD, thereby revealing potential epigenetic mechanisms. Publication III found that exposure to a feared situation induced significant changes in a plasma metabolite, namely glyoxylate, linked to the experienced anxiety level. These findings suggest that the metabolome might serve as a dynamic marker for different anxiety states. The related gene expression changes during exposure therapy were subtle but may indicate biological correlates of acute panic attacks and therapy success. Publication IV found that specific DNA methylation, such as within the gene for the serotonin receptor 3A related to the regulation of panic states and fear circuitry, as well as immune system patterns related to PD and acute fear (including arginase 1 gene methylation, CD4+ T cells, CD8+ T cells, B cells, and granulocytes), were altered during and after successful exposure therapy. Further, these changes were related to therapy response. In combination, the findings of studies III and IV highlight the possible role of epigenetics in the effects of CBT and its potential as a biomarker for treatment outcome.
In publication V, we analyzed psychological process variables throughout the course of the therapy, namely the relationship between attachment style changes and the working alliance. We found that an insecure, specifically anxious attachment style was reduced after treatment. Further, decreased anxious attachment was associated with improvements in symptom severity, therapeutic alliance, and interpersonal difficulties, which in turn were related to the effectiveness of CBT. Furthermore, findings showed that general attachment style and interpersonal distress seem to moderate the relation between the therapeutic alliance and subsequent symptom severity. The study highlights the dynamics and potential positive effects of standardized short-term CBT on interpersonal patterns.
The publications provide insights into both the pathophysiology of AD and the treatment mechanisms involved in CBT, aiming to enhance understanding of effective treatments and personalized approaches for AD patients. They emphasize, on the one hand, biological mechanisms potentially influenced by CBT: changes in the HPA system (cortisol levels), genetics (DNA methylation and gene expression), metabolomic shifts (glyoxylate regulation), and immune system dynamics. On the other hand, psychological processes are corroborated, highlighting the importance of cognitive factors, interpersonal patterns, and therapeutic alliance in the treatment of PD. Overall, multilevel changes are shown, which support a bio-psycho-social approach to PD treatment. Despite its limitations, this work highlights the probable multifaceted impact of CBT on AD/ PD and advances understanding of treatment mechanisms which could guide a more integrative approach to mental health. Future progress lies in tailoring treatments to individual needs and tracking outcomes with both biomarkers and psychological assessments.
This cumulative dissertation addresses four problems in resilient network design under disruption uncertainty with a specific focus on waterborne disruptions that affect the transportation costs within the network.
The first paper covers the integrated resilient network design problem under transportation cost un-certainty. The focus is on the integration of strategic, tactical, and operational resilience decisions, as well as the influence of the ability to predict disruptions in the short-term, and potential limitations of the operational decisions within capacity limits. The problem is modeled as a two-stage stochastic program. In addition, a problem-specific Benders decomposition algorithm is developed with a non-standard split of decision variables to solve particularly large problem instances.
The second paper focuses on the data-driven optimization of the operational inventory replenishment decision under transportation cost uncertainty. The problem is presented as a multi-stage stochastic program. To solve the problem, a cost-focused machine learning framework is proposed that uses the replenishment decisions under perfect information to train a decision tree that then learns a cost-optimal replenishment policy.
The third paper evaluates the influence of product characteristics by considering the multi-product two-echelon resilient network design problem under transportation cost uncertainty. Both different product characteristics and the dependency of considering multiple products are analyzed.
The fourth paper outlines a seven-step framework that covers the end-to-end steps a decision maker needs to follow to increase their supply chain resilience. The framework guides decision makers through the process from outlining clear resilience objectives to monitoring overall and enduring success. In addition, problem specifics and characteristics for the network design problem under waterborne disruptions are outlined to facilitate future research.
