22455 research outputs found
Sort by
In vivo Detektion von reaktiver Astrozytose und Synapsenverlust in Alzheimer-Mausmodellen mittels Positronen-Emissions-Tomographie
Get PDFDetektion Borrelia-burgdorferi-sensu-lato-spezifischer Antikörper in Serumproben geimpfter und ungeimpfter Pferde mittels drei verschiedener Line Immunoassays
Get PDFDiagnosis of equine Lyme borreliosis (LB), an infection caused by members of the Borrelia burgdorferi sensu lato complex (Bbsl), is challenging due to the nonspecific clinical signs of the disease and due to the variety of non-standardized serological tests. Specific vaccine-induced antibodies against LB, providing an effective protection against the infection, complicate the issue further. The standard for the detection of specific antibodies against Bbsl is a two-tier test system based on an enzyme-linked immunosorbent assay (ELISA) or indirect fluorescent antibody test (IFA) for antibody screening combined with a qualitative, highly specific immunoassay (e. g. line immunoassay (LIA)) for confirmation. In this study, three LIAs available for detection of antibodies in equine serum samples were evaluated and compared. A total of 393 serum samples of 131 horses with known serostatus were used. It included groups of non-vaccinated horses, immunized horses (vaccinations against LB on days 0 and 14), and horses that had received an initial immunization plus an additional booster on day 180. Sera were collected on days 0, 135 and 210 of the study. Results were compared considering the tests’ sensitivity, specificity, diagnostic outcome, and the operability of each test. Agreements of the diagnostic results among the LIAs were calculated for overall test results and single antigen-antibody-complex signal results. They are presented as inter-rater agreement and statistic reliability, represented by the Fleiss’ kappa coefficient. Agreement scores ranged from poor to moderate depending on group and time-point of blood sample collection. Depending on LIA used, deficiencies were observed in the form of non-sufficient sensitivity of antigen signals on the LIA strips (especially for outer surface protein A (OspA) or variable major protein like sequence expressed (VlsE)) or as an inappropriate test interpretation of the OspA signal. Operability of the three LIAs was equally user-friendly with minor variations. In two LIAs, test-evaluation was simplified by a supplied scanner and automatic evaluation software. To improve functionality of available LIAs for equine serum samples it is advisable to adjust sensitivity and specificity of single test antigen signals and establish appropriate evaluation protocols
Computational approaches to enhance the integrity of social media
Get PDFSocial media has profoundly transformed society. However, the proliferation of harmful content - such as hate speech, misinformation, and conspiracy theories - challenges platform integrity and poses a significant threat to society. Enhancing the integrity of social media is thus of utmost importance. In this dissertation, we propose a computational approach to improve the integrity of social media along three key dimensions: (1) the detection and understanding of harmful content, (2) audits of social media platforms, and (3) interventions to counter abusive behavior. To achieve this, we combine state-of-the-art methods from computer science with insights from the social sciences and explore each dimension through distinct case studies in three parts.
The first part focuses on detecting and understanding harmful content. Specifically, we employ a machine learning approach to identify QAnon conspiracy theorists on Parler and profile their characteristics compared to other users. We then statistically analyze the diffusion dynamics of online rumors to gain deeper insights into how harmful content spreads on social media.
The second part shifts attention to auditing social media platforms. Here, we examine how proprietary algorithms, often beyond societal control, can perpetuate biases. As a case study, we analyze the delivery of political ads on Meta during the 2021 German Federal Election.
The third part explores interventions aimed at countering harmful content and fostering civil behavior on social media. This includes results from a large-scale, pre-registered field experiment evaluating the effectiveness of AI-generated counterspeech, employing large language models, in reducing online hate speech.
By combining computational methodologies with interdisciplinary perspectives, this dissertation advances the understanding of social media vulnerabilities and delivers actionable solutions for cultivating safer digital environments. Scientifically, it demonstrates the practical application of machine learning, natural language processing, and social media analytics in detecting harmful content, auditing opaque algorithms, and evaluating scalable interventions. These contributions extend computational social science literature and inform strategies for algorithmic accountability and fair content delivery. On a societal level, the results emphasize the importance of transparency in social media platforms and highlight both the potential and risks of automated interventions, such as AI-generated counterspeech, in curbing online harms. Overall, this dissertation provides a roadmap for platform providers and policymakers committed to promoting equity, inclusivity, and democratic values within social media ecosystems.Soziale Medien haben unsere Gesellschaft tiefgreifend verändert. Die Verbreitung schädlicher Inhalte - wie Hassrede, Fehlinformationen und Verschwörungstheorien - stellt jedoch eine große Herausforderung für die Integrität von Plattformen dar und bedroht unsere Gesellschaft. Die Verbesserung der Integrität sozialer Medien ist daher von höchster Bedeutung. In dieser Dissertation schlagen wir einen computergestützten Ansatz vor, um die Integrität sozialer Medien in drei wesentlichen Bereichen zu stärken: (1) die Erkennung und das Verständnis schädlicher Inhalte, (2) Audits von Social-Media-Plattformen und (3) Interventionen zur Bekämpfung von schädlichem Verhalten. Um dies zu erreichen, kombinieren wir modernste Methoden der Informatik mit Erkenntnissen aus den Sozialwissenschaften und untersuchen jeden dieser Bereiche in separaten Fallstudien, die in drei Teilen dargestellt werden.
Der erste Teil konzentriert sich auf die Erkennung und das Verständnis schädlicher Inhalte. Konkret verwenden wir maschinelles Lernen, um QAnon-Verschwörungstheoretiker auf der Plattform Parler zu identifizieren und deren Nutzungsverhalten im Vergleich zu "normalen" Nutzern zu analysieren. Anschließend analysieren wir statistisch die Dynamik der Verbreitung von Online-Gerüchten, um tiefere Einblicke in die Verbreitung schädlicher Inhalte in sozialen Medien zu gewinnen.
Der zweite Teil beschäftigt sich mit Audits von Social-Media-Plattformen. Hier untersuchen wir, wie proprietäre Algorithmen, die sich oft der gesellschaftlichen Kontrolle entziehen, Verzerrungen verstärken können. Als Fallstudie analysieren wir die Verbreitung politischer Werbung auf Meta während der deutschen Bundestagswahl 2021.
Der dritte Teil befasst sich mit Interventionen, um schädlichen Inhalten entgegenzuwirken und respektvolles Verhalten in sozialen Medien zu fördern. Dazu erörtern wir die Ergebnisse eines groß angelegten, präregistrierten Feldexperiments, in dem die Wirksamkeit von KI-generierter Gegenrede (Counterspeech) unter Verwendung großer Sprachmodelle (Large Language Models) bei der Reduzierung von Online-Hassrede untersucht wird.
Durch die Kombination von Methoden aus der Informatik mit interdisziplinären Ansätzen erweitert diese Dissertation unser Verständnis für Schwachstellen von sozialen Medien und präsentiert Lösungen für die Schaffung eines sichereren digitalen Umfelds. Auf wissenschaftlicher Ebene demonstriert diese Arbeit die praktische Anwendung von maschinellem Lernen, natürlicher Sprachverarbeitung (Natural Language Processing) und Social-Media-Analysen zur Erkennung schädlicher Inhalte, der Prüfung von Algorithmen und der Bewertung skalierbarer Interventionen. Diese Beiträge erweitern die Computational Social Science Literatur und liefern wertvolle Erkenntnisse zum verantwortungsvollen Umgang mit Algorithmen und fairer Verbreitung von Inhalten auf Sozialen Medien. Auf gesellschaftlicher Ebene unterstreichen die Ergebnisse die Bedeutung von Transparenz auf Social-Media-Plattformen und verdeutlichen sowohl das Potenzial als auch die Risiken automatisierter Interventionen, wie z.B. KI-generierte Gegenrede, zur Eindämmung von Online-Hassrede. Insgesamt liefert diese Dissertation damit konkrete Vorschläge für Plattformbetreiber und politische Entscheidungsträger zur Förderung von Gerechtigkeit, Inklusivität und demokratischen Werten in sozialen Medien
Impact of gestational diabetes on maternal labour outcomes
Get PDFThis research explores the effects of gestational diabetes mellitus on maternal and labour outcomes, with a particular emphasis on the duration of induction, overall labour progression, and the frequency of non-elective caesarean sections. A cohort of 128 pregnant women was examined, consisting of 93 in a control group and 35 diagnosed with GDM. While maternal age and parity were similar between both groups, a notable distinction was the significantly higher pre-pregnancy BMI in the GDM group (29.7 kg/m² compared to 23.56 kg/m²). Interestingly, despite the elevated BMI, women with GDM experienced less weight gain during pregnancy, likely due to enhanced dietary regulation and closer clinical monitoring.
Labour outcomes indicated that women in the GDM group had a shorter induction phase, but their overall labour was longer, particularly during the first stage. This aligns with previous research suggesting that metabolic changes in GDM pregnancies may impact uterine contractility. However, no significant differences were observed in the duration of the second stage of labour between the two groups.
Regarding the mode of delivery, the GDM group had a significantly higher rate of caesarean sections, mainly due to an increase in planned procedures. Although the rate of unplanned caesarean sections did not differ significantly, GDM was linked to a greater likelihood of labour interventions overall. In terms of neonatal outcomes, there were no substantial differences in birth weight or Apgar scores between the groups, and the risk of macrosomia did not appear to increase in GDM pregnancies.
These findings highlight the importance of careful maternal weight management and vigilant monitoring of labour in GDM-affected pregnancies. While effective management of GDM can reduce certain neonatal risks, the longer labour durations and increased rate of caesarean sections suggest that more research is needed to optimize delivery outcomes for this population
Digitalization and the new geography of work
Get PDFThis dissertation studies how two major technology shocks – broadband Internet and remote work – reshape the spatial distribution of economic activity. In four self-contained chapters, I empirically analyze the effects of the new geography of work on urban and regional economic outcomes in Germany. Using innovative, large-scale data and state-of-the-art microeconometric methods, I provide causal evidence on four key effects. The first essay shows that high-speed broadband Internet availability significantly increases rural real estate prices, reflecting its economic value to households. However, we find that broadband subsidies aimed at closing the rural-urban divide are often fiscally ineffective. Shifting from broadband to remote work, the next three essays explore its spatial effects on cities. The second essay finds that higher WFH adoption among residents leads to reduced mobility but increased local consumer spending, indicating shifts in economic activity. The third essay examines urban housing markets, showing that WFH decreases both the price premium for central locations and spatial inequality in housing costs within cities. Finally, the fourth essay studies office real estate, finding that WFH-intensive industries downsize office space, move toward higher-quality buildings, and relocate closer to city centers. These findings highlight distinct spatial patterns: While remote work decentralizes housing demand and consumer spending, it induces a centralization effect in office real estate. This dissertation extends prior U.S.-focused research with evidence from Germany, where denser cities, stronger public transit networks, and different land-use policies create different spatial dynamics. By examining how digitalization reshapes labor markets, real estate, and economic geography, my dissertation adds to the urban and regional economics literature. The results offer insights for navigating the future of cities and labor markets in the digital age
