Publication Server of Hochschule Darmstadt
Not a member yet
259 research outputs found
Sort by
Self-supervised language models in journalism: quality perception of GPT-3-written articles
Get PDFThis experimental study investigates readers’ perceived text quality and trust towards journalistic opinion pieces written by the language model GPT-3. GPT-3 is capable of automatically writing texts in human language and is often referred to as an artificial intelligence (AI). In a 2x2x2 within- subjects experimental design, 192 participants were presented with two randomly selected articles each for evaluation. The articles were varied with regard to the variables actual source, declared source (in each case human-written or AI-written) and the topic (1 & 2). Prior to the experimental design, participants indicated the extent to which they agreed with various statements about the trustworthiness of AI in order to capture their personal attitudes towards the topic.
The study found for one, that readers considered articles written by GPT-3 to be just as good as those written by human journalists. The AI-generated versions were rated slightly better in terms of text quality as well as the trust placed in the content. However, the effect was not statistically significant. For another, no negative effect on article perception was found for texts disclosed as AI-written. Articles declared as written by an AI were mostly rated equally well or again minimally better than texts declared as human, especially regarding trust. Only the readability was rated slightly worse for the case of declaring the AI as a source. Furthermore, a correlation was found between the participants’ personal attitudes towards the topic of AI and their perception of allegedly AI-written articles. For articles declared as AI-written, there are slight to moderate positive correlations of the personal attitudes towards AI with each quality rating criterion. Personal preconception thus plays a role in the perception of AI-written articles.Die vorliegende Experimentalstudie untersucht die von LeserInnen wahrgenommene Textqualität von journalistischen Meinungsbeiträgen, die von dem Sprachmodell GPT-3 geschrieben wurden, sowie das Vertrauen in diese Texte. GPT-3 ist in der Lage, automatisch Texte in menschlicher Sprache zu schreiben und wird häufig als künstliche Intelligenz (KI) bezeichnet. In einem 2x2x2 within-subjects Experimentaldesign wurden 192 Versuchspersonen jeweils zwei zufällig ausgewählte Artikel zur Bewertung vorgelegt. Die Artikel variierten hinsichtlich der Variablen "tatsächliche Quelle", "behauptete Quelle" (jeweils Mensch oder KI) und "Thema" (1 oder 2). Vor der Versuchsdurchführung gaben die Teilnehmenden an, inwieweit sie verschiedenen Aussagen über die Vertrauenswürdigkeit von KI zustimmen, um ihre persönliche Einstellung zum Thema zu erfassen.
Die Studie ergab zum einen, dass die LeserInnen die von GPT-3 verfassten Artikel für genauso gut geschrieben befanden wie die von menschlichen JournalistInnen. Die KI-generierten Versionen wurden sowohl in Bezug auf die Textqualität als auch auf das Vertrauen in den Inhalt geringfügig besser bewertet. Der Effekt war jedoch statistisch nicht signifikant. Zum anderen wurde kein negativer Effekt auf die Qualitätsbewertung von jenen Texten festgestellt, die als von einer KI verfasst ausgegeben wurden. Angeblich KI-geschriebene Artikel wurden überwiegend gleich gut oder minimal besser bewertet als Texte, die als menschlich deklariert wurden. Lediglich die Lesbarkeit wurde in diesem Fall geringfügig schlechter bewertet. Darüber hinaus wurde ein Zusammenhang zwischen der persönlichen Einstellung der Befragten zum Thema KI und der Bewertung von vermeintlich von KI verfassten Artikeln festgestellt. Für die Bewertungen der angeblich KI-generierten Artikel fanden sich leichte bis mittlere positive Korrelationen mit den persönlichen Einstellungen zum Thema KI. Persönliche Vorurteile spielen also eine Rolle bei der Wahrnehmung von KI-geschriebenen Artikeln
Evaluation of the Explanatory Power Of Layer-wise Relevance Propagation using Adversarial Examples
Get PDFApproaches for visualizing and explaining the decision process of convolutional neural networks (CNNs) have recently received increasing attention. Particularly popular approaches are so-called saliency methods, which aim to assign a valence to each input pixel based on its importance and influence on the classification via saliency maps. In our paper, we contribute by a novel analyzing approach build on adversarial examples to investigate the explanatory power of saliency methods exemplified by layer-wise relevance propagation (LRP). Based on the hypothesis that distinct decisions, such as an image’s classification and the classification of its corresponding adversarial examples, should yield to dissimilar saliency maps to provide transparent rationales, we break down relevance scores of images and corresponding adversarial examples and analyze them using a comprehensive statistical evaluation. It turns out that different relevance decomposition rules of LRP do not lead to clearly distinguishable saliency maps for images and corresponding adversarial examples, neither in terms of their contour lines, nor in terms of the statistical analysis
Investigation of X‑band and Ka‑band amplitude scintillation based on measurement data collected during ESA’s BepiColombo superior solar conjunction campaign
Get PDFSolar phase scintillation and solar amplitude scintillation are fundamentally important in deep space mission operations for designing a communication system capable of transmitting signals when the signal path is close to the Sun. The ESA’s BepiColombo measurement data were analyzed in a previous paper in terms of the power spectral density of the solar phase scintillation, also with a comparison with Woo’s solar phase scintillation theory, when X-band and Ka-band signals propagate close to the Sun with a small Sun-Earth-Probe (SEP) angle during the superior solar conjunction campaign in March 2021 in its cruise phase to Mercury. In this paper the solar amplitude scintillation is analyzed both by calculating the power spectral density and the scintillation index. The results of scintillation index, derived from these measurement data, fit the NASA JPL’s scintillation index model
“Evaluation of an electromagnetically actuated drum brake concept”
Get PDFIn publications and conferences on the subject of wheel brakes, different concepts of electromechanically actuated wheel brakes can be found, as well as investigations into their suitability for the use in passenger cars. The vast majority of these brakes are disc or drum brakes, which are actuated by an electric motor. In the present publication, a brake concept is considered, that combines an electromagnetically actuated full-pad disc brake with a 10″ duo-duplex drum brake. The brake concept is researched in a project regarding brakes for autonomous shuttles and thus dimensioned using vehicle data of an example shuttle. The electromagnet was designed using finite element methods and the overall brake prototypically realized. The validation of the system design is carried out in component and system tests. The results show the suitability of the concept for the selected vehicle in terms of dynamics, installation space and energy requirements. However, there is a strong dependence of the braking torque output on the frictional sliding speed. Using hypothesis-based testing, electromagnetic effects like eddy currents are ruled out as a possible cause and the friction coefficient within the full-pad disc brake is identified as the main cause for the loss in torque. Consequently, the associated development conflict is identified and lies in the double function of the flux-carrying material in the electromagnet, which also acts as a friction partner for the braking disc
Combining knowledge bases for small wins in peripheral regions. An analysis of the role of innovation intermediaries in sustainability transitions
Get PDFA growing number of economic geography scholars have discussed the
spatial dimensions of sustainability transitions (STs), which entail radical changes in socio-technical systems to overcome societal, economic, and ecological problems. This involves innovation processes with a broad range of distinctive actors. Innovation intermediaries, such as universities and research institutes, are needed to support and accelerate the transfer of knowledge. Nevertheless, little is known about the influence of such actors on the configuration of the knowledge bases required for STs. This article presents insights from 14 semi-structured interviews with experts conducted in a regional innovation system (RIS) in East Germany. In cooperation with the Eberswalde University for Sustainable Development, we investigate four innovation intermediaries in the region of Eberswalde. The analytical framework links the concept of differentiated knowledge bases to small wins. Our results show that, first, in the Eberswalde region, the relevant actors involved in regional knowledge transfer focus predominantly on synthetic knowledge bases, such as experiencebased knowledge of local area settings. Second, symbolic knowledge bases are crucial and often prerequisites for intermediary organizations to recombine knowledge bases and support the capability to innovate in regional knowledge transfer. Symbolic knowledge entails the ability to translate scientific findings to a language that can be understood by the various actors in knowledge transfer. Third, changes in organizational structures complement changes in cultural–cognitive and normative institutions to support innovation on a systemic level and foster change processes
Exploring the variable importance in random forests under correlations: a general concept applied to donor organ quality in post-transplant survival
Get PDFRandom Forests are a powerful and frequently applied Machine Learning tool. The permutation variable importance (VIMP) has been proposed to improve the explainability of such a pure prediction model. It describes the expected increase in prediction error after randomly permuting a variable and disturbing its association with the outcome. However, VIMPs measure a variable’s marginal influence only, that can make its interpretation difficult or even misleading. In the present work we address the general need for improving the explainability of prediction models by exploring VIMPs in the presence of correlated variables. In particular, we propose to use a variable’s residual information for investigating if its permutation importance partially or totally originates from correlated predictors. Hypotheses tests are derived by a resampling algorithm that can further support results by providing test decisions and p-values. In simulation studies we show that the proposed test controls type I error rates. When applying the methods to a Random Forest analysis of post-transplant survival after kidney transplantation, the importance of kidney donor quality for predicting post-transplant survival is shown to be high. However, the transplant allocation policy introduces correlations with other well-known predictors, which raises the concern that the importance of kidney donor quality may simply originate from these predictors. By using the proposed method, this concern is addressed and it is demonstrated that kidney donor quality plays an important role in post-transplant survival, regardless of correlations with other predictors
Digital gestresst durch Vorgesetzte?
Get PDFTechnostress – d. h. Stress, der aus dem Umgang mit digitalen Technologien resultiert – ist eine gravierende Schattenseite der voranschreitenden Digitalisierung der Arbeitswelt. Die negativen Auswirkungen dieses Phänomens sind bereits heute erkennbar. Sie beinhalten sowohl negative gesundheitliche Folgen für die betroffenen Mitarbeiter_innen als auch gravierende Folgekosten für Unternehmen durch gesteigerte Fehlzeiten sowie negative Auswirkungen auf Mitarbeiterproduktivität und -zufriedenheit. Die vorliegende Studie untersucht, ob das Führungsverhalten einer Führungskraft die Entstehung von Technostress bei den ihr direkt unterstellten Mitarbeiter_innen beeinflusst. Darüber hinaus werden Einflüsse weiterer individueller und organisationaler Faktoren überprüft. Mittels validierter Erhebungsinstrumente werden selbstberichtete Daten von N=849 Mitarbeiter_innen deutscher Unternehmen erhoben. Die Einschätzung des Führungsverhaltens der direkten Führungskraft erfolgt auf Grundlage der Führungsstile des „Full Range of Leadership Modells“ nach Avolio und Bass (1991) unter Zuhilfenahme des MLQ 5x short. Die Ergebnisse der Datenauswertung mittels Strukturgleichungsmodellierung weisen darauf hin, dass das Führungsverhalten der bzw. des direkten Vorgesetzten Einfluss auf das Technostress-Empfinden der Mitarbeiter_innen hat.Technostress – i. e. stress resulting from dealing with digital technologies – is a serious downside to the advancing digitalization of the work environment. We are already witnessing the negative effects of this phenomenon. They include negative health consequences for the affected employees and serious follow-up costs for companies stemming from increased absenteeism and the negative effects on employee productivity and satisfaction. This study investigates whether the leadership behavior of a supervisor influences the development of technostress among the employees directly reporting to him or her. In addition, the impact of further individual and organizational factors is examined. By means of validated survey instruments, self-reported data from N=849 employees of German companies was collected. We based our assessment of the leadership behavior of the direct supervisor on the leadership styles described by the “full range of leadership model” of Avolio and Bass (1991), with the assistance of the MLQ 5x short. The results of the data evaluation by means of structural equation modeling indicate that the leadership behavior of the direct supervisor does influence the employees’ perception of technostress
Hybrid Forecasting Methods—A Systematic Review
Get PDFTime series forecasting has been performed for decades in both science and industry. The forecasting models have evolved steadily over time. Statistical methods have been used for many years and were later complemented by neural network approaches. Currently, hybrid approaches are increasingly presented, aiming to combine both methods’ advantages. These hybrid forecasting methods could lead to more accurate predictions and enhance and improve visual analytics systems for making decisions or for supporting the decision-making process. In this work, we conducted a systematic literature review using the PRISMA methodology and investigated various hybrid forecasting approaches in detail. The exact procedure for searching and filtering and the databases in which we performed the search were documented and supplemented by a PRISMA flow chart. From a total of 1435 results, we included 21 works in this review through various filtering steps and exclusion criteria. We examined these works in detail and collected the quality of the prediction results. We summarized the error values in a table to investigate whether hybrid forecasting approaches deliver better results. We concluded that all investigated hybrid forecasting methods perform better than individual ones. Based on the results of the PRISMA study, the possible applications of hybrid prediction approaches in visual analytics systems for decision making are discussed and illustrated using an exemplary visualization