52442 research outputs found
Sort by
Stakeholder strategy in action: visual, verbal, and virtual persuasion in activist hedge fund campaigns
No full textSmart start: A critical analysis of chatbots as tools to support study orientation
No full textIn higher education, elevated dropout rates pose a significant problem, leading to vast societal and personal costs and disadvantages. In Germany, in particular, dropout is closely connected to a misfit between students’ expectations and their choice of study program. As a countermeasure, a chatbot is proposed that targets prospective students and is supposed to help them with their study orientation by providing them with accurate and personalized information. In this study, we first use a survey including the Technology Acceptance Model to test whether the target group would even accept such a chatbot and whether there are differences in the acceptance for different demographic groups. Our results show high acceptance and no differences among demographic groups, although we propose that differences among demographic groups should be scrutinized in future research. Subsequently, we develop a chatbot architecture using Large Language Models and Retrieval Augmented Generation that enables accurate and personalized answers while simultaneously ensuring appropriate responses to inappropriate user prompts. Our first evaluation of the chatbot shows promising results. The chatbot will become part of a study orientation platform in one of Germany’s largest federal states
Kooperationen mit der Wissenschaft fördern neuartige Innovationen deutscher Unternehmen
No full textInnovationen sind essenziell für die Wettbewerbsfähigkeit der deutschen Wirtschaft. Kooperationen zwischen Unternehmen und wissenschaftlichen Einrichtungen gewinnen dabei zunehmend an Bedeutung. Unternehmen, die mit wissenschaftlichen Einrichtungen zusammenarbeiten, haben häufiger Erfolg bei der Einführung neuer Produkte und erzielen höhere Umsatzanteile mit diesen. Besonders stark zeigt sich dieser Effekt bei Produkten, die als Markt- oder Weltmarktneuheiten eingeführt werden. Auch der Anteil der Unternehmen, die mit wissenschaftlichen Einrichtungen kooperieren, ist gestiegen – vor allem unter jenen, die Markt- und Weltmarktneuheiten auf den Weg bringen.
Trotz der wachsenden Bedeutung solcher Kooperationen sehen sich Unternehmen mit erheblichen Herausforderungen konfrontiert. Am häufigsten geben kooperierende Unternehmen an, dass ein Mangel an öffentlicher Förderung dazu führt, dass Kooperationen entweder nicht beginnen oder vorzeitig abgebrochen werden. Zusätzlich erschweren fehlende Ressourcen vonseiten der Unternehmen oder der wissenschaftlichen Einrichtungen die Zusammenarbeit. Bereits laufende Kooperationen werden aus Sicht der Unternehmen besonders durch administrative und rechtliche Rahmenbedingungen in den wissenschaftlichen Einrichtungen behindert.
Die Ergebnisse machen deutlich, dass die finanzielle Unterstützung von Kooperationen von entscheidender Bedeutung ist, um die Innovationskraft der deutschen Wirtschaft zu stärken. Darüber hinaus würde der Abbau administrativer und rechtlicher Hürden dazu beitragen, die Zusammenarbeit zwischen Unternehmen und wissenschaftlichen Einrichtungen effektiver zu gestalten
Data Science Services an der UB Mannheim : Daten neu denken in den digitalen Geisteswissenschaften
No full textDas Forschungsdatenzentrum der Universitätsbibliothek Mannheim bietet ein breit gefächertes Angebot an digitalen Services, Plattformen und Unterstützungsformaten für die digitalen Geisteswissenschaften und im Bereich der Data Literacy. Dazu zählen spezialisierte Services wie MAObjects zur Präsentation digitaler Objektsammlungen mit Omeka sowie die automatisierte Spracherkennung mit whisply zur Unterstützung von Forschungsprojekten. Als Teil des Kompetenzzentrums OCR fördert die UB Mannheim gemeinsam mit der UB Tübingen die automatisierte Texterkennung – sowohl innerhalb der Universität als auch für externe Vorhaben. Ergänzt wird das Portfolio durch digitale Lernangebote wie MaDaLi² und durch projektbasierte Lehrinitiativen zur Förderung datenbezogener Kompetenzen. Ziel ist es, sowohl Forschende als auch Studierende im Umgang mit digitalen Methoden zu qualifizieren und die nachhaltige Verankerung digitaler Infrastrukturen in der wissenschaftlichen Praxis zu unterstützen. Der Beitrag gibt einen Überblick über das vernetzte Service- und Projektangebot der UB Mannheim, stellt ausgewählte Anwendungsszenarien vor und diskutiert zentrale Herausforderungen an der Schnittstelle von Infrastruktur, Qualifizierung und methodischer Innovation
Dance to my tune! Discovery mode and built-in recommendation bias
No full textWe examine the strategic considerations and effects of product placement services in digital content aggregators, focusing in particular on Spotify’s “Discovery Mode”. Discovery Mode introduces bias in users’ consumption bundle that content providers pay through discounted royalties, and triggers users to actively adjust to it in response. The platform’s ability to manipulate consumption bundles (and adjust users’ participation fee consistently) leads to promotion of the cheapest content available and degradation of users’ effective consumption bundles. In equilibrium, Discovery Mode can either benefit or harm the provider of the cheapest content to stream, with the harm arising whenever Discovery Mode threatens to revert preexisting bias against said provider. Importantly, Discovery Mode always forces users to costly adjust their consumption more, unequivocally generating loss of efficiency in the market. We further highlight an indirect increase in the risk of market concentration stemming from the platform’s ability to bias consumption
Who counts? Survey data quality in the age of AI
No full textLarge language models (LLMs) have been hoped to make survey research more efficient, while also improving survey data quality. However, as they are based on Internet data, LLMs may come with similar potential pitfalls as other digital data sources with regard to making inferences about human attitudes and behavior. As such, they not only have the potential to mitigate, but also to amplify existing biases regarding our understanding of different populations and constructs of interest. In this dissertation, I investigate whether and under which conditions LLMs can be leveraged in survey research by providing empirical evidence of the potentials and limits of two major applications: supplementing survey data with LLM-generated data, and coding open-ended survey responses with LLMs. I test these applications in previously unexamined contexts – European countries and languages. I conclude that LLMs cannot fully replace, but could augment human-powered survey research, given proper supervision and validation
Targeted nudging: leveraging gender differences to enhance sustainable purchase behavior, (A2025-126189)
No full textThe moral compass of politics: parties’ use of morality in political communication
No full textMorality and politics are deeply intertwined: morality provides politics with meaning, while politics gives morality institutional form and influence. This dissertation explores how moral values are expressed, contested, and strategically deployed within political discourse. Using Moral Foundations Theory (MFT) as a conceptual and analytical framework, it examines how political parties across European multi-party systems use moral language to communicate ideology, shape debate, and pursue strategic goals. MFT conceptualizes morality as a multidimensional construct grounded in five universal moral foundations: Care-Harm, Fairness-Cheating, Loyalty-Betrayal, Authority-Subversion, and Sanctity-Degradation. These moral foundations allow for systematic analysis of how moral appeals vary across ideological and contextual settings.
Methodologically, the dissertation makes a key contribution by developing and validating multilingual Moral Foundations Dictionaries in French, German, Italian, and Spanish, along with a novel cross-linguistic validation procedure for dictionary-based text analysis. Substantively, it provides the first cross-national, party-level study of moral rhetoric across European democracies. The findings demonstrate that moralized language follows predictable ideological patterns: left-wing parties emphasize individualizing foundations (Care and Fairness), while right-wing parties prioritize binding ones (Loyalty and Authority). However, the analysis also reveals that moral appeals are issue- and context-dependent, rather than purely ideology-driven. The dissertation further investigates the behavioral implications of moralized discourse within parliamentary debates, showing that while certain moral foundations (Care, Fairness, Loyalty) elicit stronger reactions, morality itself is not a primary driver of disagreement; strategic and institutional factors play a larger role.Finally, an examination of election campaign communication demonstrates that parties strategically adjust their moral rhetoric in response to electoral dynamics: adopting inclusive, balanced appeals in mass mobilization phases and reverting to partisan, binding rhetoric when consolidating core support.
Overall, this dissertation advances both methodological and theoretical understandings of moral communication in politics. It extends MFT beyond its Anglophone and individual-level applications, enriches the study of party competition and political communication, and shows that moral rhetoric is simultaneously ideological, emotional, and strategic, a central mechanism through which parties seek to persuade, mobilize, and govern
Symbolic rule-based knowledge graph completion
No full textKnowledge graphs are composed of relational facts that encode correct statements about the world. In the problem of knowledge graph completion, the goal is to augment a knowledge graph with missing but correct statements. Symbolic rule-based approaches perform this task by using logical clauses or rules. They are fully interpretable and efficient to use when using suitable inference mechanisms. A symbolic approach for performing knowledge graph completion can be separated into two stages. In the first stage, rules are learned inductively from the knowledge graph using a mining system. In the second stage, they must be applied to derive new facts. The focus of this thesis is on the latter stage which can become challenging when knowledge graphs are large and when many rules are learned. We explore and investigate how to effectively employ rules to make new fact predictions. We demonstrate in this context how rules make fully explainable predictions and how they can be used to explain other model classes. We analyse rule inference from a theoretical view and propose methods to make new predictions using supervision from a training knowledge graph. To have a meaningful evaluation, the rule-based approach and the proposed methods are compared with popular latent-based models that learn embeddings from the knowledge graph. Our results show that rule-based approaches are highly competitive when considering the predictive performance. Finally, we explore other model classes which inspires simple augmentation strategies to increase the expressiveness of the rule-based approach