Digital Library of Gesellschaft für Informatik e.V.
Not a member yet
40758 research outputs found
Sort by
Queryfy: from knowledge graphs to questions using open Large Language Models - Enabling finetuning by question generation on given knowledge
No full textWhen we look at the global knowledge graph landscape, we quickly find that there are billions of interconnected facts that have the potential to answer all kinds of questions. However, a persistent challenge lies in finding corresponding questions that align with these facts. The availability of these questions along with matching SPARQL queries is an important prerequisite for fine-tuning Large Language Models for domain-specific query generation, which is why we propose Queryfy, a novel framework that leverages Large Language Models to automate the task of deriving questions and queries from knowledge graphs, empowering users to harness their full potential
Design-time analysis of energy consumption effects using compression for mobile devices
No full textThe usefulness and acceptance of a mobile device (MD) depends among other things on the operating lifetime. Wi-Fi is one of the most familiar and highly used wireless mobile networking technologies currently available. However, Wi-Fi has a relatively high energy consumption compared to other system components. This paper describes an approach to extend the operating lifetime of a MD through power savings in the Wi-Fi communication. The approach allows for a design-time assessment if power savings are possible with lossless compression or not. It enables a system architect to specify the properties of the Wi-Fi connection and the MD and allows for estimates of the data size and expected compression ratio. These properties can be specified using well-known architectural modeling techniques in the Palladio component model
Breaking down barriers to warning technology adoption: usability and usefulness of a messenger app warning bot
No full textIn crisis situations, citizens’ situational awareness is paramount for effective response. While warning apps offer location-based alerts, their usage is relatively low. We propose a personalised messaging app channel as an alternative, presenting a warning bot that may lower adoption barriers. We employ the design science research process to define user requirements and iteratively evaluate and improve the bot’s usability and usefulness. The results showcase high usability, with over 40 % expressing an interest in utilising such a warning channel, stressing as reasons the added value of proactive warnings for personalised locations while not requiring a separate app. The derived requirements and design solutions, such as graphically enhanced user interface elements as guardrails for effective and error-free communication, demonstrate that a suitable warning chatbot does not necessarily require complex language processing capabilities. Additionally, our findings facilitate further research on accessibility via conversational design in the realm of crisis warnings
How can Design Thinking benefit Cybersecurity?: Insights from Don Norman’s The Design of Everyday Things
No full textHuman error and insufficient security awareness remain the largest cyber-risk factors for organizations. Despite the prevalence of security training, employees often fail to translate knowledge into secure behavior leading to a gap between security awareness and secure behaviour. Hence, the integration of human factors beyond awareness in cybersecurity is crucial wherein the focus lies on steering the actions executed by people rather than the technical protection offered by the security systems. Donald A. Norman’s The Design of Everyday Things is one of the pioneering books that introduces how intended actions can be achieved through a usercentric product design. Consequently, it provides a lens to rethink the various security policy designs that are developed to enforce cybersecurity. This short paper therefore proposes a new framework involving design thinking principles to help design better security policies with a human factor focus
Semantic Validation for Slingshot Simulator Using MontiArc
No full textAs software systems become increasingly complex, the demand for effective analysis has grown substantially. To meet this demand, various analysis techniques and components need to be integrated that cover the do main comprehensively. However, it is crucial to ensure that these components work together in a coherent and semantically meaningful manner. The purpose of this paper is to investigate validating the semantic correctness of the Slingshot simulator, which illustrates the composition of three distinct components of analysis. Semantic correctness in system analysis ensures that interactions and data exchanges between components accurately reflect the intended behavior and properties of the system. We use the MontiArc framework to enhance component models with constraints. It demonstrates the behavior of components using automata and verifies that these constraints are valid within automata transitions, ensuring that the automata do not produce invalid results
Tracking the Shift to Decentralised Identity: A Case Study in International Air Travel
No full textThis study explores how the air travel industry, provoked by the pandemic, shifted from a centralised paradigm towards a more decentralised one, in the form of verifiable health credentials. The study chronicles this shift, drawing on interviews with key decision–makers, participant observation and analysis of key documents. At the centre of the transition is IATA, representing the airlines, and ICAO, which sets standards for passports and digital travel credentials. The foundations for the shift began before the pandemic with the convergence of technologies (blockchain, biometrics and wallets) with new visions (frictionless travel experience, minimal handling of passenger data). The pandemic saw a travel crisis, leading to intense debate. Almost as soon as the vaccinations began, health status proof–of–concept applications were put into place using a variety of decentralised implementations. As the pandemic dissipated, the effort focused on health credentials dissipated as well, but what remained was the paradigm shift towards decentralised identities for international air travel
Investigating Quality Attributes of Machine Learning Inference on the Edge-Cloud Continuum
No full textDeploying Machine Learning (ML) tasks on the network edge is a recent research topic driven by use cases like robotics or Industry 4.0. Edge Computing is concerned with processing data close to the data source, processed by resource-constraint, heterogeneous and network-connected devices. Challenges of deploying ML tasks at edge environments include fulfilling Service Level Objectives (SLOs) like high accuracy, minimal response time, and low resource consumption. Current approaches focus on scaling the system either horizontally or vertically. In this work, we claim that offloading the ML tasks to other node types and reconfiguring the depth of the underlying ML model are additional valid actions that comply with the SLOs. We show that claim by benchmarking an image classification task utilising the ResNet model on edge hardware. Our results show that the best results are achieved when the model complexity and the hardware match each other rather than selecting the largest models or machines
From Requirements to Architecture: Semi-Automatically Generating Software Architectures
No full textTo support junior and senior architects, I propose developing a new architecture creation method that leverages LLMs’ evolving capabilities to support the architect. This method involves the architect’s close collaboration with LLM-fueled tooling over the whole process. The architect is guided through Domain Model creation, Use Case specification, architectural decisions, and architecture evaluation. While the architect can take complete control of the process and the results, and use the tooling as a building set, they can follow the intended process for maximum tooling support. The preliminary results suggest the feasibility of this process and indicate major time savings for the architect
Datensets für Requirements Engineering auf Hugging Face
No full textEs wird untersucht, ob auf der Plattform Hugging Face geeignete Datensets für das Requirements Engineering verfügbar sind. Dabei wird eine qualitative und quantitative Analyse von Datensets durchgeführt, die auf ihre Struktur, Inhalte und Konformität mit Plattformanforderungen überprüft werden. Die Ergebnisse zeigen, dass zwar einige relevante Datensets vorhanden sind, diese jedoch häufig heterogen, unvollständig oder schwer maschinell zu verarbeiten sind. Zugleich existiert mit mlm-tapt-requirements auch ein sehr umfangreiches Datenset