004 Datenverarbeitung; Informatik
Refine
Year of publication
Document Type
- Article (267)
- Doctoral Thesis (150)
- Working Paper (122)
- Bachelor Thesis (55)
- Conference Proceeding (53)
- Preprint (50)
- Diploma Thesis (49)
- Part of a Book (42)
- Contribution to a Periodical (40)
- diplomthesis (30)
Is part of the Bibliography
- no (926)
Keywords
- Lambda-Kalkül (21)
- Inklusion (13)
- Formale Semantik (11)
- Barrierefreiheit (10)
- Digitalisierung (10)
- machine learning (10)
- Operationale Semantik (9)
- artificial intelligence (9)
- data science (9)
- lambda calculus (9)
Institute
- Informatik (484)
- Informatik und Mathematik (106)
- Präsidium (75)
- Medizin (55)
- Frankfurt Institute for Advanced Studies (FIAS) (54)
- Wirtschaftswissenschaften (49)
- Physik (35)
- Hochschulrechenzentrum (24)
- studiumdigitale (24)
- Extern (12)
This dissertation investigates L1 and L2 language comprehension via translation task that accounts for “good -enough” language comprehension pioneered by Lim and Christianson (2013a,b). There are two main experiments, the first one tests native Urdu speakers with English as L2, the second experiment test native German speakers with English as L2. The nature of stimuli in both experiments consisted of matched L1 and L2 sentences with either subject relative clauses (SRCs) or object relative clauses (ORCs) – that is a first factor ‘word order canonicity’ varied whether relative clauses occurred with either canonical or non-canonical word order. All relative clauses were semantically irreversible. By reversing the arguments within the relative clause, implausible relative clauses were created, resulting in a second factor ‘Plausibility’. The two factors Word Order Canonicity and Plausibility were combined with the different translation directions (English–Urdu and Urdu–English in Experiment 1, English–German and German–English in Experiment 2) so that the combined effects of structure and meaning could be investigated for different language pairs and different directions of translation, addressing the question of whether the “good-enough” approach to language comprehension proposed by Ferreira (2003) (see Karimi and Ferreira, 2016) provides a valid framework for L1 and L2 language processing.
The World Health Organization declared the emergence of the novel coronavirus (SARS-CoV-2) in January 2020. To trace infection chains, Germany launched its smartphone contact tracing app, the “Corona-Warn-App” (CWA), in June 2020. In order to be successful as a tool for fighting the pandemic, a high adoption rate is required in the population. We analyse the respective factors influencing app adoption based on the health belief model (HBM) with a cross-sectional online study including 1752 participants from Germany. The study was conducted with a certified panel provider from the end of December 2020 to January 2021. This model is primarily known from evaluations of medical treatments, such as breast cancer screenings, but it was rarely applied in prior work for a health-related information system such as the CWA. Our results indicate that intrinsic and extrinsic motivation to use the CWA are the strongest drivers of app use. In contrast, technical barriers, privacy concerns and lower income are the main inhibitors. Our findings contribute to the literature on the adoption of contact tracing apps by questioning actual users and non-users of the CWA, and we provide valuable insights for policymakers regarding influences of adoption and potential user groups of disease prevention technologies in times of pandemics.
Background: The German Corona-Warn-App (CWA) is a contact tracing app to mitigate the spread of SARS-CoV-2. As of today, it has been downloaded approximately 45 million times.
Objective: In this study, we investigate the influence of (non-)users' social environments on the usage of the CWA during two time periods with relatively lower death rates and higher death rates caused by SARS-CoV-2.
Methods: We conducted a longitudinal survey study in Germany with 833 participants in two waves to investigate how participants perceive their peer-groups opinion about making use of the German CWA to mitigate the risk of SARS-CoV-2. In addition, we asked whether this perceived opinion, in turn, influences the participants with respect to their own decision to use the CWA. We analyzed these questions with Generalized Estimating Equations (GEE). And with two-related-samples tests to test for differences between users of the CWA and non-users and between the two points in time (wave 1 with the highest death rates observable during the pandemic in Germany versus wave 2 with significantly lower death rates). Results: Participants perceive that peer-groups have a positive opinion towards using the CWA, with more positive opinions by the media, family doctors, politicians and virologists/RKI and a lower, only slightly negative opinion originating from social
media. Users of the CWA perceive their peer groups’ opinions about using the app as more positive than non-users do. Furthermore, the perceived positive opinion of the media and politicians is significantly lower in wave 2 compared to wave 1. The perceived opinion of friends and family as well as their perceived influence towards using the CWA is significantly higher in the latter period compared to wave 1. The influence of virologists (in Germany primarily communicated via the Robert Koch Institute) has the highest positive effect on using the CWA. We only find one decreasing effect of the influence of politicians.
Conclusions: Opinions of peer groups play an important role when it comes to the adoption of the CWA. Our results show that the influence of Virologists / Robert Koch Institute and family/friends exerts the strongest effect on participants decision to use the CWA while politicians had a slightly negative influence. Our results indicate that it is crucial to accompany the introduction of such a contact tracing app with explanations and a media campaign to support its adoption which is backed up by political decision-makers subject-matter experts.
Background: The German Corona-Warn-App (CWA) is a contact tracing app to mitigate the spread of SARS-CoV-2. As of today, it has been downloaded approximately 45 million times.
Objective: This study aims to investigate the influence of (non)users’ social environments on the usage of the CWA during 2 periods with relatively lower death rates and higher death rates caused by SARS-CoV-2.
Methods: We conducted a longitudinal survey study in Germany with 833 participants in 2 waves to investigate how participants perceive their peer groups’ opinion about making use of the German CWA to mitigate the risk of SARS-CoV-2. In addition, we asked whether this perceived opinion, in turn, influences the participants with respect to their own decision to use the CWA. We analyzed these questions with generalized estimating equations. Further, 2 related sample tests were performed to test for differences between users of the CWA and nonusers and between the 2 points in time (wave 1 with the highest death rates observable during the pandemic in Germany versus wave 2 with significantly lower death rates).
Results: Participants perceived that peer groups have a positive opinion toward using the CWA, with more positive opinions by the media, family doctors, politicians, and virologists/Robert Koch Institute and a lower, only slightly negative opinion originating from social media. Users of the CWA perceived their peer groups’ opinions about using the app as more positive than nonusers do. Furthermore, the perceived positive opinion of the media (P=.001) and politicians (P<.001) was significantly lower in wave 2 compared with that in wave 1. The perceived opinion of friends and family (P<.001) as well as their perceived influence (P=.02) among nonusers toward using the CWA was significantly higher in the latter period compared with that in wave 1. The influence of virologists (in Germany primarily communicated via the Robert Koch Institute) had the highest positive effect on using the CWA (B=0.363, P<.001). We only found 1 decreasing effect of the influence of politicians (B=–0.098, P=.04).
Conclusions: Opinions of peer groups play an important role when it comes to the adoption of the CWA. Our results show that the influence of virologists/Robert Koch Institute and family/friends exerts the strongest effect on participants’ decisions to use the CWA while politicians had a slightly negative influence. Our results also indicate that it is crucial to accompany the introduction of such a contact tracing app with explanations and a media campaign to support its adoption that is backed up by political decision makers and subject matter experts.
This study compares the performance of various machine learning methods in predicting the outcomes of mergers and acquisitions (M&A), with application in merger arbitrage. Merger arbitrage capitalizes on price inefficiencies around merger announcements, empirically offering consistent, near-market-neutral returns with Sharpe ratios around 1.20 and a beta of 0.14. Leveraging logistic regression, random forest, gradient boosting machine, and neural network, I analyse 21,020 M&A deals with up to 522 predictors from 1999 to 2023. I examine two datasets: one with all features available prior to deal resolution, serving as an upper bound for predictability, and another with only features available on the announcement. Among the applied methods, XGBoost outperforms in predicting deal closure probabilities, with pseudo-out-of-sample receiver operating characteristic area under the curve (ROC-AUC) scores of 0.99 and 0.81 for the full-feature and announcement-date-only sets, respectively.
I apply these predictions to cash-only merger arbitrage from 2021 to 2023, using a classification method and testing a promising fair value investment criterion. I find that equal-weighted portfolios perform best, driven by diversification and small-size premia, achieving annualized alphas of 10 to 20% against the Fama-French five-factor model. XGBoost’s superior predictive power translates into the best merger arbitrage performance, delivering Sharpe ratios of up to 1.57 for long-only portfolios and 0.60 for zero-net-investment long-short strategies, with the latter maintaining market neutrality. I confirm these results during a second trading period from 2018 to 2020, revealing different market dynamics and similar or better model performance, with Sharpe ratios as high as 2.15.
These findings establish new benchmarks for M&A deal closure prediction, highlight the value of machine learning-driven strategies in enhancing merger arbitrage performance, and offer valuable insights for both researchers and practitioners.
With the escalating demand in the realm of natural language processing (NLP), the development of contextualized language models has become a major research focus. These models have demonstrated to be highly effective across a wide range of tasks. Nevertheless, their tokenization methodologies mostly rely on statistical measures, not taking linguistic knowledge into account. This thesis investigates the idea of directly injecting morphological knowledge into contextualized language models without the need for a full pre-training. I introduce MoVE (Morphological Vocabulary Extension), an approach that employs a database of morphological segmentations to enhance the initial tokenization. Additionally, fine-tuning the model's word embeddings to directly incorporate morphological information into the token embeddings. The proposed method is evaluated on a morphologically challenging task-created through the extraction of complex words from Amazon reviews-in both English and German, showing that it is able to outperform the baseline model.
The study of animal behavior is essential for gaining a better understanding of the behavior, patterns, and needs of animals. A better understanding not only serves scientific progress, but also plays an important role in improving husbandry conditions in zoos, which can help to improve animal welfare (Berger, 2010; Brando and Buchanan-Smith, 2018; Walsh et al., 2019; Rose and Riley, 2021).
The behavior of large herbivores differs significantly between day and night, and most ungulates are diurnal or crepuscular (Bennie et al., 2014; Gravett et al., 2017; Davimes et al., 2018; Wu et al., 2018). In contrast, many studies examine animal behavior during the day, and unfortunately there is little information on nocturnal behavior, including sleep behavior (Berger, 2010; Rose and Robert, 2013). However, sleep behavior, especially the proportion of REM sleep, is of great importance for the well-being of an individual (Hänninen et al., 2008; Fukasawa et al., 2018; Northeast et al., 2020).
To gain more insight into the behavior of ungulates in general, studies based on large samples of different species with a long recording period are useful. This is difficult to achieve with manual data analysis, as data collection and analysis in behavioral biology is time consuming and costly. Therefore, modern methods such as automated analysis are helpful in the field of behavioral biology (Norouzzadeh et al., 2018; Beery et al., 2020; Lürig et al., 2021).
Hence, the development of a software tool for the automated assessment of nocturnal behavior of ungulates in zoos is part of this dissertation. The resulting software tool is called BOVIDS (Behavioral Observations by Videos and Images using Deep-Learning Software) and allows the automatic evaluation of video material in three steps. In the first step, object detection, the individuals on the images are recognized and cut out in order to classify the behavior in the following step, action classification. In the final step, post-processing, errors of the automated analysis are corrected and the data is prepared for further use (Hahn-Klimroth et al., 2021; Gübert et al., 2022). To create such a system, it must first be trained. Typically, two nights per individual were manually annotated, resulting in a total of 594 manually annotated nights. In addition, 224,922 images were used to evaluate whether the system was already correctly recognizing the animals' behavior. Bounding boxes were either manually drawn or evaluated on a total of 201,827 images to train the object detection network.
The software package BOVIDS was used to analyze data from a total of 196 individuals from 19 different ungulate species over a period of 101,629 hours of video material from 9,239 nights. A night is defined as the period from 7 pm to 6 am. The species studied belong to the two orders of odd-toed ungulates (Perissodactyla) and even-toed ungulates (Artiodactyla). The focus is on the behavioral categories of standing, moving, lying – head up, and lying – head down, the latter corresponding to the typical REM sleep position of ungulates. Based on the analyzed data, several biological questions were discussed in this thesis. In addition to the activity budgets and rhythms underlying the night, factors influencing behavior are also investigated. In addition, the enclosure use by the animals is evaluated.
Zebras as representatives of the Perissodactyla spend about 25% of the night lying, while the average for the Artiodactyla studied is 77%. All species studied spend an average of 8.8% of the night in REM sleep (Gübert et al., 2023a), with a typical REM sleep phase lasting between 2.2 and 7.6 minutes (Gübert et al., 2023b). Only 0.7% of time during the night is spent with movement by the animals (Gübert and Dierkes). While the number of lying phases within the Artiodactyla is very constant with an average of five phases, the number of phases in the REM sleep position varies. Age, average species size and taxonomy were found to be influencing factors (Gübert et al., 2023a). With regard to rhythmicity, it is striking that most of the species studied show an increase in lying during the night and that a strong rhythmicity of behavior can be observed. The time between two lying events is very constant and is about two hours for most animals (Gübert et al., 2023b). With regard to enclosure use, it is striking that only a small part of the enclosure is used regularly. All individuals prefer to lie down on the bedding and most individuals prefer one or two different resting places (Gübert and Dierkes).
The data created as part of this thesis can contribute to a better overall understanding of ungulate behavior. The newly developed software package BOVIDS makes it relatively easy to analyze further data on this topic. Long-term studies can now be carried out more cost-effectively, making it easier to answer many questions in the future, such as investigating other influencing factors or responses to changes in husbandry conditions.
Delegated online search
(2024)
In a delegation problem, a principal P with commitment power tries to pick one out of 𝑛 options. Each option is drawn independently from a known distribution. Instead of inspecting the options herself, P delegates the information acquisition to a rational and self-interested agent A. After inspection, A proposes one of the options, and P can accept or reject. Delegation is a classic setting in economic information design with many prominent applications, but the computational problems are only poorly understood. In this paper, we study a natural online variant of delegation, in which the agent searches through the options in an online fashion. For each option, he has to irrevocably decide if he wants to propose the current option or discard it, before seeing information on the next option(s). How can we design algorithms for P that approximate the utility of her best option in hindsight? We show that in general P can obtain a Θ(1∕𝑛)-approximation and extend this result to ratios of Θ(𝑘∕𝑛) in case (1) A has a lookahead of 𝑘 rounds, or (2) A can propose up to 𝑘 different options. We provide fine-grained bounds independent of 𝑛 based on three parameters. If the ratio of maximum and minimum utility for A is bounded by a factor 𝛼, we obtain an Ω(loglog 𝛼∕ log 𝛼)- approximation algorithm, and we show that this is best possible. Additionally, if P cannot distinguish options with the same value for herself, we show that ratios polynomial in 1∕𝛼 cannot be avoided. If there are at most 𝛽 different utility values for A, we show a Θ(1∕𝛽)-approximation. If the utilities of P and A for each option are related by a factor 𝛾, we obtain an Ω(1∕ log 𝛾)- approximation, where 𝑂(log log 𝛾∕ log 𝛾) is best possible.
The expanding field of epitranscriptomics might rival the epigenome in the diversity of biological processes impacted. In recent years, the development of new high-throughput experimental and computational techniques has been a key driving force in discovering the properties of RNA modifications. Machine learning applications, such as for classification, clustering or de novo identification, have been critical in these advances. Nonetheless, various challenges remain before the full potential of machine learning for epitranscriptomics can be leveraged. In this review, we provide a comprehensive survey of machine learning methods to detect RNA modifications using diverse input data sources. We describe strategies to train and test machine learning methods and to encode and interpret features that are relevant for epitranscriptomics. Finally, we identify some of the current challenges and open questions about RNA modification analysis, including the ambiguity in predicting RNA modifications in transcript isoforms or in single nucleotides, or the lack of complete ground truth sets to test RNA modifications. We believe this review will inspire and benefit the rapidly developing field of epitranscriptomics in addressing the current limitations through the effective use of machine learning.