Digital Library of Gesellschaft für Informatik e.V.
Not a member yet
40758 research outputs found
Sort by
Für eine goldene Mitte
No full textDer akademische Mittelbau in Deutschland bildet das Rückgrat von Forschung und Lehre an Universitäten – aber wie zufrieden sind diejenigen, die diese zentrale Rolle ausfüllen? Durch befristete Verträge, hohes Arbeitspensum und unsichere Karriereperspektiven stehen viele von ihnen vor großen Herausforderungen. Der Beirat Junge Wissenschaft der GI hat sich mit ihrer Lage genauer auseinandergesetzt
Key parameters to increase the adoption of AI in agricultural decision support
No full textThis research paper contributes to the field of the application of artificial intelligence (AI) in German viticulture. The research gap addressed in this paper is the lack of research on the adoption of AI-based decision support in viticultural businesses. Consequently, expert interviews were conducted, yielding a multifaceted understanding of winegrowers' attitudes towards AI. The generated insights provide five dimensions with 48 subcategories that can influence the adoption of an AI system into a winery. From these subcategories, 14 parameters can be derived within the viticultural company, as well as external pressures from the environment and the AI system itself. This work leads to two further findings, which can be promoted as follows: firstly, the dimension “Functional & Physical Design of AI” can be derived from the interviews; secondly, the deductively derived “Trust in AI” is confirmed as another dimension alongside the dimensions described by the underlying framework
Scalable Computation of Shapley Additive Explanations
No full textThe growing field of explainable AI (XAI) develops methods that help better understand ML model predictions. While SHapley Additive exPlanations (SHAP) is a widely-used, model-agnostic method for explaining predictions, its use comes with a significant computational burden, particularly for complex models and large datasets with many features. The key—and so far unaddressed—challenge lies in efficiently scaling these computations without compromising accuracy. In this paper, we present a scalable, model-agnostic SHAP sampling framework on top of Apache SystemDS. We leverage Antithetic Permutation Sampling for its efficiency and optimization potential, and we devise a carefully vectorized and parallelized implementation for local and distributed operations. Compared with the state-of-the-art Python shap package, our solutions yield similar accuracy but achieve significant speedups of up to 14x for multi-threaded singlenode operations as well as up to 35x for distributed Spark operations (on a small 8 node cluster)
A Classification Framework for Scientific Documents to Support Knowledge Graph Population
No full textResearch papers are a central communication medium to share new scientific insights and progress and are, nowadays, stored as PDF files. However, little effort is spent on reorganizing information with effective knowledge classification and comprehensive representation during the publication process. In terms of software engineering (SE), those papers are also aligned to software artifacts and research data. Aggregating knowledge and empirical evidence is done with systematic literature studies that tend to be very time-consuming and require a manual inspection of the respective research artifacts (paper-and data-wise). Research knowledge graphs like the Open Research Knowledge Graph (ORKG) aim to contribute to and rethink scholarly communication by providing formats while easing the processing of semantic information. Therefore, ORKG offers templates to summarize and structure a research artifact’s content, providing metadata as well. Based on this, researchers can connect similar papers, reuse replication artifacts, and generate literature studies more easily. However, adding papers to ORKG is still a tedious manual process. Moreover, selecting suitable template formats is challenging and can be highly domain-specific. To support the manual process, we aim to provide an automated classification framework for scientific papers, supporting the knowledge graph population. This framework intends to be flexible, allowing various input data and schemas, so it can be applied and trained in a multitude of research fields in SE. In this paper, we present the concept of the framework’s implementation, and an excerpt of the evaluation result in one of the research subfields in SE, namely software architecture and design
Exploring Data Mesh Architecture: A Comparative Study of Implementation Archetypes Across Different Sectors and Industries
No full textAn emerging idea in data architecture and platform thinking, called Data Mesh, has been gaining significant attention in recent years. This paper delves into the concept of Data Mesh and examines industry implementations. The research identifies three different architectural patterns that are applied depending on the level of domain agility of the organization. By evaluating semi-structured interviews with practitioners, the study highlights the factors that influence the successful implementation of these patterns and offers insights into their experiences. These findings enhance the comprehension of Data Mesh implementation and the architectural patterns that support it and can aid organizations in adopting Data Mesh by providing valuable guidance for practitioners and researchers
Häufige Fehler und mögliche Fehlvorstellungen von Schüler:innen der 10. Klasse. Eine Analyse offener Aufgaben
No full textDas Wissen um häufige Fehler und Fehlvorstellungen von Schüler:innen ermöglicht einen individualisierten, auf das Vorwissen der Schüler:innen angepassten Unterricht. Es wurde eine Analyse von Antworten eines Fachwissenstests von 404 Schüler:innen () am Ende der 10. Klasse durchgeführt und mittels qualitativer Inhaltsanalyse ausgewertet. Häufige Fehler und mögliche Fehlvorstellungen der Schüler:innen werden im Poster aufgezeigt
Gemeinsam Lernen leicht gemacht: Vernetzung mit dem Buddy Finder. Entwicklung eines Matching-Systems für Studierende auf Grundlage einer Nutzerstudie.
No full textWie können sich Lernende in einer zunehmend digitalen Bildungswelt sinnvoll vernetzen? Im BMBF-Projekt „Bildungsraum Digital“ entstand mit dem „Buddy Finder“ ein Tool, welches insbesondere Studierende, aber auch andere Lernende anonym zusammenbringt. Die Matching-Kriterien wurden mithilfe von Kriterienkatalogen verwandter Vernetzungstools konzipiert und anschließend durch eine Online-Umfrage mit 686 Teilnehmenden empirisch validiert. Die Ergebnisse flossen in einen neuen Kriterienkatalog und einen interaktiven Prototyp ein. Der „Buddy Finder“ stellt ein Fundament zur Verbesserung der digitalen Vernetzung und Förderung kollaborativen Lernens im Bildungsbereich dar
From Risks to Routes: Human-Centered Development of a Multi-Platform Planning Support System for Safer School Commutes
No full textThe daily school commute plays a crucial role in developing children’s ability to travel independently. In Germany, school route plans, which are maps that highlight safe paths and danger zones, are a well-established way of improving the safety of school commutes. However, over half of municipalities lack such plans, often due to limited resources and the significant coordination effort required. In this paper, we present a human-centered approach to developing a multi-platform planning support system designed to facilitate the creation of school route plans. Through low-fidelity prototyping and iterative user research involving planners, parents, and mobility experts, we identified key functional requirements. The resulting software prototype includes a mobile app for data collection and a desktop application for plan creation. This paper details the user research, derived requirements and the implementation of the system. Our approach shows how digital tools can support local authorities and lower barriers to participation for parents and children
Entrepreneurship im Ehrenamt
No full textDie Redaktion bat mich, einen Text beizusteuern, um unserer ehemaligen Sprecherin Edna Kropp zu danken. Dieser Aufforderung komme ich gerne nach. Mein Text hat eine persönliche Note, denn Edna und ich sind ein eingespieltes Team. Wir haben gemeinsam die GI-Frauen Regionalgruppe Berlin aufgebaut und insbesondere während der Corona-Pandemie intensiv bespielt