Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
14 results
Search Results
Item Restricted Artificial intelligence for bias detection in higher education online content(Saudi Digital Library, 2026) Bin Shiha, Rawan; Eric, Atwell; Noorhan, AbbasThis thesis develops and evaluates Artificial Intelligence (AI) and Natural Language Processing (NLP) approaches for detecting bias in higher education online resources, addressing the lack of systematic computational methods in this area. Bias in higher education content shapes knowledge production, representation, and equity, making its detection both academically and socially significant. To address this gap, three sets of novel datasets were created: 1. a corpus of university news articles annotated for subjectivity, sentiment, and gender representation; 2. three university reading list datasets with demographic annotations enabling comparative analysis across Western and Middle Eastern contexts; and 3. two domain-specific collections of learning materials, one from humanities-oriented open resources and the other from Science, Technology, Engineering, and Mathematics (STEM) lecture transcripts. Together, these datasets provide the first systematic resources for investigating representational, stereotypical and linguistic bias across diverse higher education domains. Using these resources, a range of Pre-trained Language Models (PLMs) and Large Language Models (LLMs) were evaluated for bias detection. PLM revealed significant gendered and representational disparities in university discourse, while fine-tuned LLM achieved improved performance on humanities data but showed limited transferability to STEM materials. A hybrid framework integrating fine-tuned LLMs with Retrieval-Augmented Generation (RAG) enhanced detection transparency and prediction balance. These methods were operationalised in a prototype web application for bias detection in higher education learning content. The contributions of this research are fourfold: 1. the creation of multiple novel annotated datasets spanning university news, reading lists, and academic learning resources; 2. the introduction of LLM-based strategies for bias annotation, fine-tuning, and cross-domain evaluation 3. methodological innovations including a structured framework for categorising bias, a hybrid human–AI annotation approach, and a replicable NLP pipeline for demographic and thematic analysis; and 4. the design of a hybrid bias detection system and accompanying web application. This work advances computational approaches to bias detection, provides reproducible resources for future research, and offers practical tools to support greater fairness and equity in higher education online environments.3 0Item Restricted NLP-Based Cybersecurity Threat Intelligence Management Framework In Manufacturing Sector: A Framework For Predicting Cyberattacks And Security Incidents - Toward The Transformation Of Industry 4.0.(Saudi Digital Library, 2026) Albarrak, Majed; Sandeep, Jagtap; Konstantinos, SalonitisIndustry 4.0 manufacturing environments depend on interconnected cyber-physical systems, Industrial Control Systems (ICS), and IoT technologies, significantly increasing exposure to sophisticated cyber threats. Modern attacks are faster, increasingly malware-free, and often detected too late by traditional signature-based defences. Early indicators of such threats frequently appear in unstructured open-source text, including social media and technical forums, long before formal advisories are issued. However, existing Cyber Threat Intelligence (CTI) practices in manufacturing make limited use of these early signals, restricting proactive defence. This limitation arises because CTI systems predominantly rely on structured and validated intelligence feeds, while early threat indicators are embedded in large volumes of noisy, unstructured textual data that are difficult to process automatically. Furthermore, extracting actionable intelligence from such sources requires complex NLP pipelines and high-quality labelled data, which remain scarce, making the timely and accurate integration of these signals into standardised frameworks (e.g., MITRE ATT&CK) challenging (Büchel et al., 2025; Rahman, Hezaveh and Williams, 2023). This research adopts a design science approach to develop and validate an NLP-based CTI management framework tailored to Industry 4.0. The framework integrates three components: (i) an urgency-aware topic modelling approach (U-BERTopic) to detect emerging threats from unstructured text streams, (ii) a transformer-based classification model (AC_MAPPER) that automatically maps extracted threat narratives to MITRE ATT&CK techniques using class-aware augmentation, and (iii) an end-to-end CTI framework embedding these models across the CTI lifecycle. Quantitative evaluation was conducted using standard NLP and classification metrics across five publicly available cyber threat intelligence (CTI) datasets: TRAM, TRAM Bootstrap, CAPEC, TTPHunter, and HALdata. These datasets consist of sentence-level annotations linking unstructured threat intelligence text to MITRE ATT&CK techniques. Their sizes vary significantly, ranging from 803 sentences in HALdata to 12,945 sentences in CAPEC, with intermediate datasets including TRAM (5,089 sentences), TRAM Bootstrap (11,130 sentences), and TTPHunter (8,887 sentences). The datasets also differ in label diversity, covering between 46 and 188 ATT&CK techniques, and exhibit substantial class imbalance, with some techniques occurring frequently (e.g., T1027, T1059) while many others appear fewer than ten times, reflecting real-world CTI data challenges. The findings show that urgency-aware topic modelling improves early identification of high-risk threat signals, with U-BERTopic achieving a higher topic diversity score of up to 0.88, compared to 0.56 for LDA, indicating more distinct and informative cybersecurity topics. Furthermore, AC_MAPPER demonstrates robust performance in mapping CTI text to ATT&CK techniques, achieving a macro F1-score of 0.83 on the CAPEC dataset, significantly outperforming baseline models such as PRIMUS (0.72) and CTI-BERT (0.65). Expert evaluation confirms that the integrated framework enhances interpretability, traceability, and practical relevance for industrial cybersecurity operations. This thesis demonstrates that NLP and Large Language Models can effectively transform unstructured textual intelligence into structured, ATT&CK-aligned CTI for manufacturing systems. The proposed framework supports earlier threat awareness and proactive defence, advancing industrial cybersecurity from reactive response toward predictive and intelligence-driven protection in Industry 4.0.13 0Item Restricted Toward Transparent, Reliable, and Trustworthy GenAI: A Framework for Grounding LLM Applications—Evidence from the Healthcare Sector in the Kingdom of Saudi Arabia(Saudi Digital Library, 2026) Altalhi, Amal; Itamar, ShabatiGenerative artificial intelligence (GenAI) is a rapidly advancing technology with growing relevance to digital healthcare and intelligent clinical decision support. However, a notable risk of large language models (LLMs) is hallucination, in which AI-generated outputs are fabricated, unsupported, or inaccurate. In emergency department triage, hallucinations may compromise patient safety, reduce clinician trust, and impede adoption. Prior research has predominantly focused on technical mitigation approaches, with relatively limited attention to sociotechnical and trust-related implementation barriers in high-stakes clinical settings. This dissertation addresses hallucination-resilient healthcare GenAI through the design and evaluation of a grounded framework intended to improve reliability, transparency, trustworthiness, and clinician adoption. To this end, the research employed a Design Science Research (DSR) methodology. The resulting artifact, the Collaborative Hallucination-Resilient Framework (CHRF), was instantiated in Trier, an AI-powered triage assistant aligned with the Canadian Triage and Acuity Scale (CTAS) and designed to support emergency department triage decision-making. The framework incorporated a hybrid processing pipeline that combined advanced Retrieval-Augmented Generation (RAG) with machine learning through an ML-first, RAG-conditional architecture to improve triage reliability, mitigate hallucinations, and support supervised clinician involvement. Trier was developed and evaluated across two iterative versions: Proposed V2 and Enhanced V2. The system was evaluated through complementary technical performance and user perception evaluations. Technical performance evaluations were conducted through iterative design cycles using 4,063 historical CTAS emergency triage cases obtained from the King Abdullah International Medical Research Center. Baseline V1, Proposed V2, Enhanced V2, and clinician benchmarks were evaluated using overall accuracy, mean absolute error (MAE), and triage deviation analysis. Baseline V1 achieved 38.00% accuracy with an MAE of 0.67, whereas Proposed V2 achieved 56.39% accuracy with an MAE of 0.45. The final Enhanced V2 model attained 83.39% accuracy and an MAE of 0.17, while substantially mitigating severe multi-level triage deviations and improving discrimination between Level 3 and Level 4 acuity labels compared with Proposed V2 and Baseline V1. Clinician accuracy remained superior, with 97.96% accuracy and an MAE of 0.02. Survey responses from 121 healthcare professionals working in five hospitals in Saudi Arabia were analyzed using partial least squares structural equation modeling (PLS-SEM), fuzzy-set qualitative comparative analysis (fsQCA), necessary condition analysis (NCA), and thematic content analysis. The PLS-SEM results revealed that Trust Propensity (p < .001), Familiarity (p < .05), and System Performance (p < .05) significantly predicted Cognitive Trust, whereas System Understanding was not significant (p > .05). In addition, Cognitive Trust significantly predicted Attitude toward Use (p < .001) and directly predicted Switching Intention (p < .001). Moreover, Attitude toward Use significantly predicted Switching Intention (p < .001) and mediated the effect of Cognitive Trust on Switching Intention (p < .001). The asymmetric analyses revealed that Cognitive Trust and Attitude toward Use were necessary conditions for high levels of Switching Intention. These findings suggest that responsible AI adoption is relational, configurational, and context-dependent rather than strictly linear. This dissertation contributes theoretically to trustworthy AI research, healthcare GenAI applications, Design Science Research, and AI adoption theory by positioning CHRF as a reusable nascent design framework for hallucination-resilient LLM applications in high-stakes healthcare environments. The study contributes not only an instantiated artifact but also reusable design knowledge, empirically grounded trust mechanisms, and an integrated sociotechnical explanation of clinician adoption behavior. From a managerial perspective, the findings provide actionable guidance for implementing evidence-grounded, uncertainty-aware, and clinician-supervised GenAI triage systems capable of supporting emergency decision-making while maintaining human clinical accountability. Acknowledged limitations include the use of a small CTAS-specific dataset with class imbalances, geographically similar participant samples, and the lack of prospective real-time clinical deployment. Future work should address these limitations by evaluating AI triage systems across broader institutions and regions using larger and more class-balanced multi-institutional datasets, better calibrating boundaries for underrepresented CTAS levels, and performing prospective real-time clinical testing. Ultimately, this dissertation shows that GenAI can responsibly assist emergency triage when its outputs are evidence-grounded, uncertainty-aware, and supervised by human clinical insight. Under these conditions, GenAI can facilitate trustworthy human-AI interaction and decision-making in one of the most time-critical healthcare domains.11 0Item Restricted Making Linked Data Discoverable In The Context Of Wildlife Data Observatories(Saudi Digital Library, 2025) Mussa, Omar; Perera, Charith; Rana, Omer; Goossens, Benoit; Orozco-terWengel, PabloIn recent years, Linked Data (LD) and Semantic Web technologies have gained traction as powerful frameworks for integrating and querying distributed datasets across disciplines. Despite their potential, the complexity of technologies such as RDF triplestores and SPARQL query languages continues to hinder adoption among non-expert users, particularly within bioscience and wildlife research, where observational data is prevalent. This thesis addresses the usability gap by exploring how LD technologies can be made more accessible for domain specialists without technical expertise in Semantic Web technologies. A mixed-methods approach was adopted, combining a systematic literature review, ethnographic fieldwork, and iterative interface design. Findings highlighted key limitations in existing LD access techniques, particularly their inability to support observational data patterns and spatiotemporal querying needs. In response, this research presents an integrated approach that combines graphical and conversational user interfaces to assist domain specialists in constructing queries without prior technical knowledge. This approach enables users to formulate complex semantic queries either through visual interactions or by describing their information needs in natural language, with the system translating these into executable queries, supporting key features that include dynamic filter generation, spatial selection, and ontology-aware entity linking. The approach was evaluated through a task-based user study involving bioscience researchers. Results demonstrate that the integrated interface significantly improves usability, task accuracy, task completion time (over 50% improvement across most tasks) and user satisfaction compared to graphical or conversational UIs. Furthermore, this thesis explores the integration of Large Language Models (LLMs) via a Retrieval-Augmented Generation (RAG) approach to enhance semantic interpretation and user support. Across multiple use cases, integrating LLMs enhances the expressivity of natural language queries, allowing previously unsupported queries to be answered with over 89% accuracy and up to 100% for many tasks. Overall, this research contributes to the field by introducing accessible LD retrieval methods for non-experts in ecological data observatories.30 0Item Restricted How Large Language Models are Reshaping Skills and Job Requirements for Public Health Professionals in Saudi Arabia(Saudi Digital Library, 2025) Alkhinjar, Mulfi; Palmer, PaulaContext: Large Language Models (LLMs) such as ChatGPT, Gemini, and DeepSeek are transforming professional work across sectors by enhancing information processing and decision support. In public health, these technologies offer the potential to improve efficiency, analytical capacity, and data-driven decision-making. Yet, their integration raises concerns about workforce preparedness, evolving skill requirements, and ethical oversight. In Saudi Arabia, where Vision 2030 prioritizes digital transformation in healthcare, understanding how public health professionals adapt to these technologies is vital for workforce and policy planning. Method: This exploratory mixed-methods study examined the professional impact of LLMs and the preparedness of public health professionals for their integration. The validated Shinners Artificial Intelligence Perception (SHAIP) survey, adapted for LLMs and public health, was distributed to employees of the Saudi Public Health Authority, yielding 32 complete responses. Ten semi-structured interviews further explored four constructs: professional impact, preparedness, new essential skills, and obsolete skills. Quantitative data were analyzed descriptively, and qualitative data were coded using thematic analysis. Findings: Survey results indicated that LLMs positively influence efficiency and workflow but revealed gaps in training and ethical guidance. Interview themes reinforced these findings, identifying new essential skills such as prompt engineering, digital literacy, and critical oversight, while traditional tasks like manual data entry and report drafting were viewed as increasingly automated. Conclusion: LLMs are transforming the roles of public health professionals. Successful adoption requires structured training, institutional readiness, and ethical governance. The study offers actionable recommendations to align workforce development and recruitment strategies with Saudi Vision 2030, emphasizing capacity building and responsible AI integration in public health practice.21 0Item Restricted The Impact of LLMs Usage on Learning Outcomes for Software Development Students: A Focus on Prompt Engineering(Saudi Digital Library, 2025) Aljohani, Mohammed; Itamar Shabtai; June K. Hilton; Chinazunwa UwaomaThis study investigates the impact of large language model (LLM) usage, specifically ChatGPT, on student learning outcomes in programming education. The research adopts a mixed-methods approach, combining quantitative survey data from students and qualitative interviews with instructors. The study addresses three research questions: (1) the effect of LLM usage on undergraduate students' learning outcomes, (2) the influence of prompt engineering skills on this relationship, and (3) instructors' perceptions on these relationships. Quantitative data were collected from 159 students across two Saudi universities using a structured online survey with sections covering demographic information, LLM usage, self-reported programming understanding, and prompt engineering skills. Qualitative data were obtained through semi-structured interviews with programming instructors, covering LLM usage, prompt engineering skills, and their impact on student learning outcomes. The quantitative analysis utilized Partial Least Squares Structural Equation Modeling (PLS-SEM) to assess the measurement and structural models, including path coefficients, model explanatory power (R²), and predictive power (PLSpredict). Qualitative data were thematically analyzed using Atlas.ti to identify key themes related to instructor perspectives on the model. LLM usage positively impacts learning outcomes. While quantitative results did not show a significant moderating effect of prompt engineering skills, qualitative findings highlight its critical role in determining the positive effect of LLM usage on learning outcomes. The study emphasizes the importance of clear LLM usage policies and early prompt engineering training to promote meaningful engagement and maintain academic integrity in programming courses.15 0Item Restricted Improving Feature Location in Source Code via Large Language Model-Based Descriptive Annotations(Arizona State University, 2025-05) Alneif, Sultan; Alhindawi, NouhFeature location is a crucial task in software maintenance, aiding developers in identifying the precise segments of code responsible for specific functionalities. Traditional feature location methods, such as grep and static analysis, often result in high false-positive rates and inadequate ranking accuracy, increasing developer effort and reducing productivity. Information Retrieval (IR) techniques like Latent Semantic Indexing (LSI) have improved precision and recall but still struggle with lexical mismatches and semantic ambiguities. This research introduces an innovative method to enhance feature location by augmenting source code corpora with descriptive annotations generated by Large Language Models (LLMs), specifically Code Llama. The enriched corpora provide deeper semantic contexts, improving the alignment between developer queries and relevant source code components. Empirical evaluations were conducted on two open-source systems, HippoDraw and Qt, using standard IR performance metrics: precision, recall, First Relevant Position (FRP), and Last Relevant Position (LRP). Results showed significant performance gains; a 40% precision improvement in HippoDraw, and a 26% improvement in Qt. Recall improved by 32% in HippoDraw and 24% in Qt. The findings highlight the efficacy of incorporating LLM-generated annotations, significantly reducing developer effort and enhancing software comprehension and maintainability. This research provides a practical and scalable solution for software maintenance and evolution tasks.22 0Item Restricted Adversarial Machine Learning: Safeguarding Al models from Attacks(Lancaster University, 2025-01-10) Alammar, Ghaida; Bilal, MuhammadThe field of AML has gained considerable popularity over the years with researchers seeking to explore gaps and new opportunities for growth. This goal of this report is to offer an in-depth survey of adversary attacks and defences in machine learning by examining existing gaps in current algorithms and understanding the implications for systems. By exploring evasion, poisoning, extraction, and inference attacks, the paper reveals the weaknesses of the existing methodologies such as adversarial training, data sanitization, and differential privacy. These techniques are usually not versatile to newer threats and have raised concerns about how effective they are in practical use. The research contributes to the field by conducting an extensive literature review of 35 articles and highlighting the need to implement adaptive and diverse defence strategies as well as empirical studies to evaluate the effectiveness of AML mechanisms. Some of the strategic suggestions are to incorporate continuous training frameworks, optimise real-time monitoring processes, and improve privacy-preserving methods to safeguard confidential information. This analysis is envisaged to offer practical data to foster the development of AML to help in the development of robust AI systems that will remain impregnable to various kinds of adversarial threats in numerous vital sectors. The study examines the basic design and consequences of various attacks in addition to the impact of subtle manipulation of input data on patterns and privacy. The report further addresses the modern challenges of large language models (LLMs) and autonomous systems. Furthermore, this research emphasises the significance of robust protection against enemy attack in strategic areas. The studies additionally evaluate present day protection mechanisms inclusive of antagonistic training, enter preprocessing, and making models stronger and more reliable. By evaluating the efficiency of these defences and evaluating key areas for improvement, the dissertation provides invaluable insights into enhancing the security and reliability of systems. The results of addressing the attacks and defences expose the need for unremitting advancements in data protection in various systems.32 0Item Restricted Automating the Formulation of Competency Questions in Ontology Engineering(University of Liverpool, 2025) Alharbi, Reham; Tamma, Valentina; Grasso, Floriana; Payne, TerryOntology reuse is a fundamental aspect of ontology development, ensuring that new ontologies align with established models to facilitate seamless integration and interoperability across systems. Despite decades of research promoting ontology reuse, practical solutions for semi-automatically assessing the suitability of candidate ontologies remain limited. A key challenge is the lack of explicit requirement representations that allow for meaningful comparisons between ontologies. Competency Questions (CQs) , which define functional requirements in the form of natural language questions, offer a promising means of evaluating ontology reuse potential. However, in practice, CQs are often not published alongside their ontology, making it difficult to assess whether an existing ontology aligns with new requirements, ultimately hindering reuse. This thesis tackles the challenge of ontology reuse by introducing an automated approach to retrofitting CQs into existing ontologies. Leveraging Generative AI, specifically Large Language Models (LLMs), this approach generates CQs from ontological statements, enabling the systematic extraction of functional requirements even when they were not explicitly documented. The performance of both open-source and closed-source LLMs is evaluated, with key parameters such as prompt specificity and temperature explored to control hallucinations and improve the quality of retrofitted CQs. Results indicate high recall and stability, demonstrating that CQs can be reliably retrofitted and aligned with an ontology’s intended design. However, precision varies due to long-tail data effects, and potential data leakage may artificially inflate recall, necessitating further research. By enabling the reconstruction of CQs, this approach provides a foundation for assessing ontology reuse based on requirement similarity. Specifically, CQ similarity can serve as an indicator of how well an existing ontology aligns with the needs of a new ontology development effort. To operationalize this idea, this thesis proposes a reuse recommendation phase within ontology development methodologies. This phase systematically identifies candidate ontologies based on requirement overlap, offering a structured approach to reuse assessment. The methodology is validated through a practical case study, demonstrating its effectiveness in real-world ontology design. By embedding an explicit reuse recommendation step in the ontology engineering process, this approach provides ontology engineers with a systematic method to identify suitable candidate ontologies, enhancing the overall design process.28 0Item Restricted Evaluating Chess Moves by Analysing Sentiments in Teaching Textbooks(the University of Manchester, 2025) Alrdahi, Haifa Saleh T; Batista-navarro, RizaThe rules of playing chess are simple to comprehend, and yet it is challenging to make accurate decisions in the game. Hence, chess lends itself well to the development of an artificial intelligence (AI) system that simulates real-life problems, such as in decision-making processes. Learning chess strategies has been widely investigated, with most studies focused on learning from previous games using search algorithms. Chess textbooks encapsulate grandmaster knowledge, which explains playing strategies. This thesis investigates three research questions on the possibility of unlocking hidden knowledge in chess teaching textbooks. Firstly, we contribute to the chess domain with a new heterogeneous chess dataset “LEAP”, consists of structured data that represents the environment “board state”, and unstructured data that represent explanation of strategic moves. Additionally, we build a larger unstructured synthetic chess dataset to improve large language models familiarity with the chess teaching context. With the LEAP dataset, we examined the characteristics of chess teaching textbooks and the challenges of using such a data source for training Natural Language (NL)-based chess agent. We show by empirical experiments that following the common approach of sentence-level evaluation of moves are not insightful. Secondly, we observed that chess teaching textbooks are focused on explanation of the move’s outcome for both players alongside discussing multiple moves in one sentence, which confused the models in move evaluation. To address this, we introduce an auxiliary task by using verb phrase-level to evaluate the individual moves. Furthermore, we show by empirical experiments the usefulness of adopting the Aspect-based Sentiment Analysis (ABSA) approach as an evaluation method of chess moves expressed in free-text. With this, we have developed a fine-grained annotation and a small-scale dataset for the chess-ABSA domain “ASSESS”. Finally we examined the performance of a fine-tuned LLM encoder model for chess-ABSA and showed that the performance of the model for evaluating chess moves is comparable to scores obtained from a chess engine, Stockfish. Thirdly, we developed an instruction-based explanation framework, using prompt engineering with zero-shot learning to generate an explanation text of the move outcome. The framework also used a chess ABSA decoder model that uses an instructions format and evaluated its performance on the ASSESS dataset, which shows an overall improvement performance. Finally, we evaluate the performance of the framework and discuss the possibilities and current challenges of generating large-scale unstructured data for the chess, and the effect on the chess-ABSA decoder model.12 0