Refine
Year of publication
Document Type
- Doctoral Thesis (511)
- Part of Periodical (376)
- Bachelor Thesis (275)
- Master's Thesis (203)
- Diploma Thesis (196)
- Study Thesis (138)
- Article (24)
- Conference Proceeding (12)
- Other (8)
- Report (8)
Language
- German (1181)
- English (580)
- Multiple languages (4)
- Spanish (2)
Keywords
- Vorlesungsverzeichnis (63)
- Bildverarbeitung (16)
- Augmented Reality (15)
- Simulation (12)
- Computersimulation (11)
- Computergraphik (10)
- Pestizid (10)
- Robotik (10)
- Computergrafik (9)
- Computervisualistik (9)
Institute
- Institut für Computervisualistik (338)
- Fachbereich 4 (299)
- Zentrale Einrichtungen (191)
- Institut für Informatik (153)
- Institut für Wirtschafts- und Verwaltungsinformatik (153)
- Institut für Management (119)
- Fachbereich 7 (99)
- Institute for Web Science and Technologies (65)
- Institut für Softwaretechnik (55)
- Fachbereich 8 (47)
The integration of Large Language Models (LLMs) into information retrieval sys tems has transformed the user experience by providing direct, conversational responses instead of traditional ranked lists of search results. This modification raises substantial concerns about user trust, behaviour, and the risk of misinformation, even as it improves accessibility and convenience. This thesis investigates the impact of generative information retrieval on the reliability of synthesized answers, particularly focusing on how hallucination rates and semantic drift influence trust dynamics and information-seeking behavior. By evaluating the performance of different LLMs on fact-checking benchmarks, the study seeks to quantify the advantages of model scaling against the inherent risks of factual inaccuracy.
The study evaluates hallucination and user trust in LLM-augmented information retrieval systems using three fact-checking datasets. Three well-known semantic similarity metrics are employed to assess the alignment between LLM responses and ground-truth references. Furthermore, the hallucination rate and factual consistency are assessed by aligning model-generated responses with verified annotations in fact-checking datasets. We utilise bias detection measures to evaluate implicit stereotype reinforcement in LLM outputs. This study applies a comprehensive framework for evaluating and auditing hallucinations by combining quantitative performance metrics with user-level reliability insights. The work aims to establish a baseline for the transparency and reliability of LLMs in search and retrieval contexts.
Affordable RGB-D cameras have gained broad research attention in computer vision. This imaging modality provides registered 3-D and RGB data of the perceived environment in the point cloud data format. RGB-D cameras and point clouds allow for a seamless integration of object recognition in mobile robotics applications. Not only do point clouds pose a potential for novel algorithms, the provided geometric information also facilitates estimating adequate positions for mobile manipulation – an action that often follows the recognition process in mobile robotics. Research on vision algorithms based on 3-D data is far less advanced than on 2-D images. Further, RGB-D cameras also pose some technical challenges. Due to their inner workings, they provide erroneous measurements, have a limited range and a much lower resolution than RGB cameras.
In this thesis, I will focus on the object classification and detection parts of vision pipelines for mobile robotics. The main research question is whether point clouds from low-cost RGB-D cameras can be used to reliably solve these tasks. I decided to use a traditional algorithm for my investigations. Traditional algorithms usually demand less computational resources for training and require less data to build a model for inference. For these and other reasons some application scenarios might restrict or prohibit using methods based on deep neural networks (DNN).
The presented point cloud processing pipeline is a non-parametric approach to object classification and detection. It is inspired by the local Naive-Bayes Nearest Neighbor (NBNN) and the Implicit Shape Model (ISM) algorithms originally introduced for 2-D images. Local feature descriptors are used to construct a spatial code-book during the training stage. In the test stage this codebook is used in a Hough voting scheme to generate object hypotheses. I will carefully adapt ideas from the above methods and extend several pipeline steps by novel contributions. At all times, I will target fast processing with limited computational resources in mind.
In a first step, I will focus on isolated objects for classification to gain insights into using point clouds for vision. Subsequently, the presented pipeline will be extended to handle noisy data for object detection in cluttered environments. The contributions of this thesis include an efficient sampling method to find suitable locations for local descriptors and the creation of a descriptive codebook with ranked feature descriptors. The hypothesis generation is followed by an elaborate hypothesis verification step and an additional verification with global feature descriptors in an ensemble classifier. Further, I introduce modifications to two popular local descriptors and also extend them to the global scale.
The developed approaches are evaluated on publicly available datasets with simulated and real sensor data. Further, the mobile service robot Lisa was used for evaluation during several competitions and achieved excellent results. The results of this work enable fast and reliable shape classification of isolated objects, as well as object detection in cluttered environments. The complete pipeline is open-source and is published online in a software repository under a permissive license.
Along with the increased use of automation processes in every other task, job recommendations as well as hiring have also become partially automated. In the process of applying only to those jobs that are recommended by a system or choosing only from those candidates that are selected by an automated system, it becomes highly important to find out if the automated systems are trustworthy enough to provide fair decisions. There has been research on fairness in this sector, where the main focus has been on a single protected attribute, in most cases ’gender.’ That is why the aim of this research is to delve deeper into a deep learning transformer-based algorithm used for job matching with a text-based resume dataset containing several demographic attributes to investigate the fairness of the algorithm not only for gender but also for other demographic groups such as race, age group, and experience level. The fairness evaluation has been carried out using multiple fairness metrics, including demographic parity, conditional demographic parity, and equal opportunity. The transformer models that are pretrained language models are chosen for this study due to their ability to understand the meaning and context of words in resumes and job descriptions. The thesis further investigates bias using an alternative approach by working on a dataset containing varying protected attributes and then conducting a comparable analysis of several bias mitigation methods, including multiple layers of data resampling, along with sensitivity testing through data modification in the preprocessing step of the recommendation process. The research reveals that a system performing fairly when considering a single protected attribute can even hide intersectional unfairness and that the bias mitigation methods do not ensure a balanced improvement across each subgroup when considering multiple combinational demographic groups.
Die vorliegende Dissertation unternimmt eine grundlegende Neubewertung des Werkes von Stefan Andres (1906–1970), der nach seiner Zeit als Bestsellerautor der Nachkriegsjahre weitgehend aus dem literarischen Kanon verschwunden ist. Entgegen der herkömmlichen Rezeption, die Andres oft auf seinen als nicht mehr zeitgemäß erachteten christlichen Humanismus oder auf seinen Ruf als Heimatdichter reduziert, weist die Arbeit nach, dass sein Œuvre eine hohe Modernität besitzt und komplexe Identitätsdiskurse sowie Alteritätserfahrungen verhandelt, die für die globalisierte Gesellschaft des 21. Jahrhunderts von hoher Relevanz sind.
Methodisches Rückgrat der Untersuchung ist die kontrapunktische Lektüre im Sinne Edward Saids, die durch das poststrukturalistisch geprägte Instrumentarium Homi K. Bhabhas erweitert wird. Im Zentrum stehen dabei Bhabhas Konzepte der Hybridität, des Dritten Raumes, der Mimikry und der Ambivalenz. Diese Theorien, die ursprünglich im Kontext kolonialer Machtverhältnisse entstanden sind, werden in dieser Arbeit erfolgreich auf die literarische Darstellung innereuropäischer und geschlechtsspezifischer Machtasymmetrien transferiert.
Anhand der Romane „Die unsichtbare Mauer“ (1934) und „Die Reise nach Portiuncula“ (1954) wird aufgezeigt, wie Andres binäre Oppositionen – etwa zwischen urbanem Zentrum und ruraler Peripherie – sowohl inszeniert als auch subversiv unterlaufen werden. In der Analyse des Romans „Die unsichtbaren Mauer“ wird der Bau der Dhrontalsperre als eine koloniale Situation gedeutet, in der technokratisches Wissen zur epistemischen Gewalt gegenüber einer marginalisierten Landbevölkerung wird. Beim Roman „Die Reise nach Portiuncula“
wird die metaphorische Afrikanisierung des italienischen Schauplatzes demaskiert, die dazu dient, die Protagonistin Assunta als exotisierte „Andere“ zu konstruieren und patriarchale Herrschaftsansprüche zu legitimieren. Die Arbeit belegt, dass die weiblichen Figuren bei Andres sich starren Stereotypen entziehen und sich durch Formen der „schlauen Höflichkeit“ oder gewaltsamen Widerstand zu handlungsstarken Subjekten entwickeln.
Abschließend zeigt die Dissertation, dass das Re-Reading von Stefan Andres durch die Brille der Postcolonial Studies nicht nur dessen Werk neu erschließt, sondern auch die Leistungsfähigkeit kulturtheoretischer Analysewerkzeuge für die Germanistik unter Beweis stellt. Damit leistet die Arbeit einen wesentlichen Beitrag zur aktuellen Debatte um die Erweiterung des literarischen Kanons und die Dekonstruktion eurozentrischer Wahrnehmungsmuster.
In order to support both the provision of clean drinking water and the preservation of biodiversity in aquatic ecosystems, a comprehensive scientific understanding of the identity, concentration, and behavior of anthropogenic pollutants in the aquatic environment is essential. Organic micropollutants constitute a large group of anthropogenic pollutants and originate from all areas of daily human life: pharmaceuticals are part of the daily routine for many people, pesticides are crucial for food production, and organic chemicals are used in the industrial production of paper, plastics, paints, and many other products. These micropollutants enter the water cycle, either in their parent form or as transformation products, where they can lead to potentially harmful effects. In addition to target methods, non-target approaches have been established as powerful tools for comprehensively investigating these compounds in the water cycle. Beyond the analytical challenges of instrumentally detecting these compounds, the prioritization and evaluation of the large datasets generated by non-target pose both chemical and data science challenges, forming the overarching theme of this work.
The work comprises three studies demonstrating the development and application of non-target screening (NTS) methodologies. In an NTS using high-performance liquid chromatography (HPLC) coupled with high-resolution mass spectrometry (QTOF-MS/MS), 112 samples from the river Nidda and seven of its tributaries were analyzed. On average, approximately 2700 signals, or features, were detected per sample. To filter these extensive data and prioritize unknown compounds, a method was first developed to reliably assign adducts, isotopologues, source fragments and other ionization products to a common component based on retention time and peak shape. In the next step, the prioritization of unknown compounds was achieved by highlighting features that were detected specifically at individual sites under investigation, but which were not typically considered to originate from municipal wastewater. This was accomplished by comparing the data from the Nidda river system with NTS data from municipal wastewater treatment plant effluents. Only the highlighted, Nidda-specific features were considered further. As a result, an average data prioritization of 7% across all samples was achieved, leading to the identification of nine compounds. Among these were the industrial compound Nylostab S-EED™, which had not been previously observed in the environment and three algal toxins, whose occurrence resulted from the algal bloom of a nearby water body.
In the second study, investigations focused on permanently cationic compounds in suspended particulate matter samples from the rivers Rhine and Saar. The data prioritization was based on the specific physicochemical properties of this substance group. Following extraction, a two-step procedure was applied, relying on interactions with strong ion exchangers and chromatography using deuterated solvents. This resulted in 5% of the detected NTS signals being labeled as potentially cationic compounds. Based on this, 22 compounds were identified, four of which were previously unknown. Trend analyses covering the period from 2005/2006 to 2018, along with an assessment of the ecotoxicological risks based on semi-quantitatively determined concentrations, suggest that identified compounds such as Basic Yellow 28 and Fluorescent Brightener 363 may have a high ecotoxicological relevance.
The third study focused on the analysis of samples from disconnected, inter-regional river systems. In collaboration with the local environmental authorities, a three-year study was conducted, analyzing 524 samples from 79 sites along rivers across Saxony. For this purpose, a method was developed that allowed for the characterization of sites based on five categories, considering both known and unknown compounds to assess their chemical contamination. The results were classified by calculating the modified z-scores within each category. The method was validated based on the results of target analysis in the same samples. As a result of the study, 13 sites were classified as anomalous due to high z-scores, as elevated levels of pharmaceuticals, industrial chemicals but also unclassified unknown compounds were detected. At two potentially industrially contaminated sites, the Münzbach and Dorfbach Oberschindmaas, nine compounds were identified and their concentrations were estimated using a 1-point calibration. For the compound hexa(methoxymethyl)melamine, a concentration in the range of 300 µg/L was determined, suggesting a considerable risk for the environment (risk quotient: 5.6).
Microplastics (MP) (1 μm to 5 mm) enter riverine environments in various ways. Point sources (e. g. industrial plants) can be differentiated from diffuse sources (e. g. erosion). Once MP has entered river systems, it underlies hydrodynamic processes and is finally transported towards the oceans. Due to fragmentation, MP particles decrease in size during transport. MP particles are highly heterogeneous considering density and shape, which makes spatially and temporally high resolved sampling difficult and thus hampers the quantification of annual loads in rivers. Within this thesis, we conducted a spatially and temporally high resolved monitoring in the River Rhine with the aim to quantify an annual MP load at our study site, Koblenz.
With a literature review we exposed that most studies present their results by means of particle number instead of particle mass. Also, many studies did not consider spatial nor temporal aspects within their sampling design. For the first time, we presented a compilation of global annual load values within river systems, partly using innovative methodological approaches to convert MP items to MP mass. Hereby, a correlation between catchment area and annual MP load could be detected. Global loads showed a large range (< 1 kg/y – 1533 t/y).
To estimate the annual MP load of the River Rhine at the study site Koblenz, we performed a temporally high resolved sampling using sedimentation boxes (SB, time-integrated samples over months) and a continuous flow centrifuge (CFC, discrete sampling). Via thermal degradation procedures, polymer concentration in mg/g was estimated for PP, PE, PS and PVC. Hereby, annual loads of 400 t (SB) and 60 t (CFC) were calculated. Deviation of both methods was more pronounced for polymers with lower density (PP and PE), whereas PVC loads were comparable. This led to further laboratory experiments, which showed that particle retention efficiency of the SB was varying with the particle’s properties. Thus, the SB sample might not be representative for the river’s conditions. Furthermore, the CFC had difficulties in retaining polymers with lower density (losses up to > 50 % were noticed). Nevertheless, the calculated MP loads do fit well into our compilation of global MP load values. An extensive quantification of retention efficiencies for several polymer types and particle sizes (SB) or an additional sampling step for the clearwater of the CFC for low-dense MP particles could lead to more robust MP concentration values in the future.
Depth-distributed sampling campaigns with a filter cascade and filter nets revealed strong density-dependant vertical gradients within the water column. These gradients were even apparent within the smallest particle size classes and thus were larger than a theoretical determination based on the Rouse number would have suggested. Both, filter cascade and filter net showed a positive correlation between river discharge and MP concentration. This correlation allows for a first speculation on MP sources within the River Rhine: The main source of MP seems, in congruence with suspended sediments, to be a diffuse entry which rises with rising water level. In contrast, a constant entry from a point source would lead to lower concentrations during flood events due to dilution.
Within this work we could point out that extensive research is needed considering the understanding of different sampling devices and their sampling efficiencies. First approaches for an improvement of these limitations have been shown. Still, a combination of sampling techniques is necessary to cover the portfolio of potential issues regarding MP research. For a more precise quantification of annual MP loads, standardisation of procedures on all working steps is urgently needed.