Refine
Year of publication
Document Type
- Master's Thesis (16) (remove)
Keywords
- API analysis (1)
- API-Analyse (1)
- CSCA (1)
- Cicero (1)
- Computer Supported Cooperative Work (1)
- Datenschutz (1)
- Developer profiling (1)
- Echtzeit-Interaktion (1)
- Entwickler Profil (1)
- InstaHub (1)
Institute
- Institut für Informatik (16) (remove)
Assessing ChatGPT’s Performance in Analyzing Students’ Sentiments: A Case Study in Course Feedback
(2024)
The emergence of large language models (LLMs) like ChatGPT has impacted fields such as education, transforming natural language processing (NLP) tasks like sentiment analysis. Transformers form the foundation of LLMs, with BERT, XLNet, and GPT as key examples. ChatGPT, developed by OpenAI, is a state-of-the-art model and its ability in natural language tasks makes it a potential tool in sentiment analysis. This thesis reviews current sentiment analysis methods and examines ChatGPT’s ability to analyze sentiments across three labels (Negative, Neutral, Positive) and five labels (Very Negative, Negative, Neutral, Positive, Very Positive) on a dataset of student course reviews. Its performance is compared with fine tuned state-of-the-art models like BERT, XLNet, bart-large-mnli, and RoBERTa-large-mnli using quantitative metrics. With the help of 7 prompting techniques which are ways to instruct ChatGPT, this work also analyzed how well it understands complex linguistic nuances in the given texts using qualitative metrics. BERT and XLNet outperform ChatGPT mainly due to their bidirectional nature, which allows them to understand the full context of a sentence, not just left to right. This, combined with fine-tuning, helps them capture patterns and nuances better. ChatGPT, as a general purpose, open-domain model, processes text unidirectionally, which can limit its context understanding. Despite this, ChatGPT performed comparably to XLNet and BERT in three-label scenarios and outperformed others. Fine-tuned models excelled in five label cases. Moreover, it has shown impressive knowledge of the language. Chain-of-Thought (CoT) was the most effective technique for prompting with step by step instructions. ChatGPT showed promising performance in correctness, consistency, relevance, and robustness, except for detecting Irony. As education evolves with diverse learning environments, effective feedback analysis becomes increasingly valuable. Addressing ChatGPT’s limitations and leveraging its strengths could enhance personalized learning through better sentiment analysis.
Exploring Academic Perspectives: Sentiments and Discourse on ChatGPT Adoption in Higher Education
(2024)
Artificial intelligence (AI) is becoming more widely used in a number of industries, including in the field of education. Applications of artificial intelligence (AI) are becoming crucial for schools and universities, whether for automated evaluation, smart educational systems, individualized learning, or staff support. ChatGPT, anAI-based chatbot, offers coherent and helpful replies based on analyzing large volumes of data. Integrating ChatGPT, a sophisticated Natural Language Processing (NLP) tool developed by OpenAI, into higher education has sparked significant interest and debate. Since the technology is already adapted by many students and teachers, this study delves into analyzing the sentiments expressed on university websites regarding ChatGPT integration into education by creating a comprehensive sentiment analysis framework using Hierarchical Residual RSigELU Attention Network (HR-RAN). The proposed framework addresses several challenges in sentiment analysis, such as capturing fine-grained sentiment nuances, including contextual information, and handling complex language expressions in university review data. The methodology involves several steps, including data collection from various educational websites, blogs, and news platforms. The data is preprocessed to handle emoticons, URLs, and tags and then, detect and remove sarcastic text using the eXtreme Learning Hyperband Network (XLHN). Sentences are then grouped based on similarity and topics are modeled using the Non-negative Term-Document Matrix Factorization (NTDMF) approach. Features, such as lexico-semantic, lexico structural, and numerical features are extracted. Dependency parsing and coreference resolution are performed to analyze grammatical structures and understand semantic relationships. Word embedding uses the Word2Vec model to capture semantic relationships between words. The preprocessed text and extracted features are inputted into the HR-RAN classifier to categorize sentiments as positive, negative, or neutral. The sentiment analysis results indicate that 74.8% of the sentiments towards ChatGPT in higher education are neutral, 21.5% are positive, and only 3.7% are negative. This suggests a predominant neutrality among users, with a significant portion expressing positive views and a very small percentage holding negative opinions. Additionally, the analysis reveals regional variations, with Canada showing the highest number of sentiments, predominantly neutral, followed by Germany, the UK, and the USA. The sentiment analysis results are evaluated based on various metrics, such as accuracy, precision, recall, F-measure, and specificity. Results indicate that the proposed framework outperforms conventional sentiment analysis models. The HR-RAN technique achieved a precision of 98.98%, recall of 99.23%, F-measure of 99.10%, accuracy of 98.88%, and specificity of 98.31%. Additionally, word clouds are generated to visually represent the most common terms within positive, neutral, and negative sentiments, providing a clear and immediate understanding of the key themes in the data. These findings can inform educators, administrators, and developers about the benefits and challenges of integrating ChatGPT into educational
settings, guiding improvements in educational practices and AI tool development.
Modern software projects are composed of several software languages, software technologies and different kind of artifacts. Therefore, the understanding of the software project at hand, including the semantic links between the different parts, becomes a difficult challenge for a developer. One approach to attack this issue is to document the software project with the help of a linguistic architecture. This kind of architecture can be described with the help of the MegaL ontology. A remaining challenge is the creation of it since it requires different kind of skills. Therefore, this paper proposes an approach for the automatic extraction of a linguistic architecture. The open source framework Apache Jena, which is focusing on semantic web technologies like RDF and OWL, is used to define custom rules that are capable to infer new knowledge based on the defined or already extracted RDF triples. The complete approach is tested in a case study on ten different open source projects. The aim of the case study is to extract a linguistic architecture that is describing the use of Hibernate in the selected projects. In the end, the result is evaluated with the help of different metrics. The evaluation is performed with the help of an internal and external approach.
Using semantic data from general-purpose programming languages does not provide the unified experience one would want for such an application. Static error checking is lacking, especially with regards to static typing of the data. Based on the previous work of λ-DL, which integrates semantic queries and concepts as types into a typed λ-calculus, this work takes its ideas a step further to meld them into a real-world programming language. This thesis explores how λ-DL's features can be extended and integrated into an existing language, researches an appropriate extension mechanism and produces Semantics4J, a JastAdd-based Java language semantic data extension for type-safe OWL programming, together with examples of its usage.
Motion capture refers to the process of capturing, processing and trans- lating real motions onto a 3D model. Not only in the movie and gaming industries, motion capture creates an indispensable realism of human and animal movement. Also in the context of robotics, medical movement therapy, as well as in AR and VR, motion capture is used extensively. In addition to the well established optical processes, especially in the last three areas, alternative systems based on inertial navigation (IMUs) are being used in-creasingly, because they do not rely on external cameras and thus limit the area of movement considerably less.
Fast evolving technical progress in the manufacturing of such IMUs allows building small sensors, wearable on the body which can transfer movements to a computer. The development of applying inertial systems to a motion capture context, however, is still at an early state. Problems like drift can currently only be minimized by adding additional hardware for correcting the read data.
In the following master thesis an IMU based motion capture system is designed and constructed. This contains the assembly of the hardware components as well as processing of the received movement data on the software side and their application to a 3D model.
This thesis proposes the use of MSR (Mining Software Repositories) techniques to identify software developers with exclusive expertise about specific APIs and programming domains in software repositories. A pilot Tool for finding such
“Islands of Knowledge” in Node.js projects is presented and applied in a case study to the 180 most popular npm packages. It is found that on average each package has 2.3 Islands of Knowledge, which is possibly explained by the finding that npm packages tend to have only one main contributor. In a survey, the maintainers of 50 packages are contacted and asked for opinions on the results produced by the Tool. Together with their responses, this thesis reports on experiences made with the pilot Tool and how future iterations could produce even more accurate statements about programming expertise distribution in developer teams.
This thesis analyzes the online attention towards scientists and their research topics. The studies compare the attention dynamics towards the winners of important scientific prizes with scientists who did not receive a prize. Web signals such as Wikipedia page views, Wikipedia edits, and Google Trends were used as a proxy for online attention. One study focused on the time between the creation of the article about a scientist and their research topics. It was discovered that articles about research topics were created closer to the articles of prize winners than to scientists who did not receive a prize. One possible explanation could be that the research topics are more closely related to the scientist who got an award. This supports that scientists who received the prize introduced the topics to the public. Another study considered the public attention trends towards the related research topics before and after a page of a scientist was created. It was observed that after a page about a scientist was created, research topics of prize winners received more attention than the topics of scientists who did not receive a prize. Furthermore, it was demonstrated that Nobel Prize winners get a lower amount of attention before receiving the prize than the potential nominees from the list of Citation Laureates of Thompson Reuters. Also, their popularity is going down faster after receiving it. It was also shown that it is difficult to predict the prize winners based on the attention dynamics towards them.
The aim of this thesis was to develop and to evaluate a method, which enables the utilization of traditional dialog marketing tools through the web. For this purpose, a prototype of a website with "extended real-time interaction (eEI)" capabilities has been implemented and tested. The prototype was evaluated by a methodology based on the five-dimensional "e-service quality" measure after Gwo-Guang Lee und Hsiu-Fen Lin. The Foundation of the "e-service quality" measure is the SERVQUAL-Model. A statistical analysis of the user study results showed a significant correlation between eEI and user satisfaction. Before the actual realization of eEI, the "Technology Acceptance Model" after Fred D. Davis was used to investigate currently used real-time interaction systems.
The publication of open source software aims to support the reuse, the distribution and the general utilization of software. This can only be enabled by the correct usage of open source software licenses. Therefore associations provide a multitude of open source software licenses with different features, of which a developer can choose, to regulate the interaction with his software. Those licenses are the core theme of this thesis.
After an extensive literature research, two general research questions are elaborated in detail. First, a license usage analysis of licenses in the open source sector is applied, to identify current trends and statistics. This includes questions concerning the distribution of licenses, the consistency in their usage, their association over a period of time and their publication.
Afterwards the recommendation of licenses for specific projects is investigated. Therefore, a recommendation logic is presented, which includes several influences on a suitable license choice, to generate an at most applicable recommendation. Besides the exact features of a license of which a user can choose, different methods of ranking the recommendation results are proposed. This is based on the examination of the current situation of open source licensing and license suggestion. Finally, the logic is evaluated on the exemplary use-case of the 101companies project.
One task of executives and project managers in IT companies or departments is to hire suitable developers and to assign them to suitable problems. In this paper, we propose a new technique that directly leverages previous work experience of developers in a systematic manner. Existing evidence for developer expertise based on the version history of existing projects is analyzed. More specifically, we analyze the commits to a repository in terms of affected API usage. On these grounds, we associate APIs with developers and thus we assess API experience of developers. In transitive closure, we also assess programming domain experience.
Geographic cluster based routing in ad-hoc wireless sensor networks is a current field of research. Various algorithms to route in wireless ad-hoc networks based on position information already exist. Among them algorithms that use the traditional beaconing approach as well as algorithms that work beaconless (no information about the environment is required besides the own position and the destination). Geographic cluster based routing with guaranteed message delivery can be carried out on overlay graphs as well. Until now the required planar overlay graphs are not being constructed reactively.
This thesis proposes a reactive algorithm, the Beaconless Cluster Based Planarization (BCBP) algorithm, which constructs a planar overlay graph and noticeably reduces the number of messages required for that. Based on an algorithm for cluster based planarization it beaconlessly constructs a planar overlay graph in an unit disk graph (UDG). An UDG is a model for a wireless network in which every participant has the same sending radius. Evaluation of the algorithm shows it to be more efficient than the non beaconless variant. Another result of this thesis is the Beaconless LLRAP (BLLRAP) algorithm, for which planarity but not continued connectivity could be proven.
Software systems are often developed as a set of variants to meet diverse requirements. Two common approaches to this are "clone-and-owning" and software product lines. Both approaches have advantages and disadvantages. In previous work we and collaborators proposed an idea which combines both approaches to manage variants, similarities, and cloning by using a virtual platform and cloning-related operators.
In this thesis, we present an approach for aggregating essential metadata to enable a propagate operator, which implements a form of change propagation. For this we have developed a system to annotate code similarities which were extracted throughout the history of a software repository. The annotations express similarity maintenance tasks, which can then either be executed automatically by propagate or have to be performed manually by the user. In this work we outline the automated metadata extraction process and the system for annotating similarities; we explain how the implemented system can be integrated into the workflow of an existing version control system (Git); and, finally, we present a case study using the 101haskell corpus of variants.
Cicero ist eine asynchrone Diskussionsplattform, die im Rahmen der Arbeitsgruppe Informationssysteme und Semantic Web (ISWeb) der Universität Koblenz-Landau entwickelt wurde. Die webbasierte Anwendung folgt dem Gedanken eines semantischen Wikis und soll insbesondere beim Arbeitsablauf von Entwurfsprozessen eingesetzt werden. Dabei verwendet Cicero ein restriktives Argumentationsmodell, das einerseits strukturierte Diskussionen von schwierigen Prozessen fördert und andererseits den Entscheidungsfindungsprozess unterstützt. Im Zentrum der Arbeit steht die Evaluation von Cicero, wobei im vorhergehenden theoretischen Teil die Hintergründe und Funktionsweisen vorgestellt werden und im nachfolgenden praktischen Teil die Anwendung anhand einer Fallstudie evaluiert wird. Die Studie wurde im Rahmen der Übungsveranstaltung zu Grundlagen der Datenbanken der Universität Koblenz im Wintersemester 2008/2009 durchgeführt , und die Studenten hatten die Aufgabe, einen Entwurfsprozess mit Hilfe von Cicero zu diskutieren. Über die erhobenen Daten der Fallstudie wird ein Akzeptanztest durchgeführt. Hierbei wird überprüft, ob die Benutzer Cicero positiv annehmen und die Methodik richtig anwenden. Denn aufgrund des vorgegebenen Argumentationsmodells müssen die Benutzer ihr Kommunikationsverhalten ändern und ihren herkömmlichen Diskussionsstil der Anwendung anpassen. Ziel der Evaluation ist es, kritische Erfolgsfaktoren im Umgang mit Cicero ausfindig zu machen. Anhand der identifizierten Schwachstellen werden abschließend gezielte Maßnahmen vorgeschlagen, die die Akzeptanz der Benutzer gegenüber Cicero erhöhen könnten.
Problems in the analysis of requirements often lead to failures when developing software systems. This problem is nowadays being faced by requirements engineering. The early involvement of all kinds of stakeholders in the development of such a system and a structured process to elicitate and analyse requirements have made it a crucial factor as a first step in software development. The increasing complexity of modern softwaresystems though leads to a rising amount of information which has to be dealt with during analysis. Without the support of appropriate tools this would be almost impossible to do. Especially in bigger projects, which tend to be spatially distributed, an effective requirements engineering could not be implemented without this kind of support. Today there is a wide range of tools dealing with this matter. They have been in use since some time now and, in their most recent versions, realize the most important aspects of requirements engineering. Within the scope of this thesis some of these tools will be analysed, focussing on both the major functionalities concerning the management of requirements and the repository of these tools. The results of this analyis will be integrated into a reference model.