1. Towards a sustainable cybersecurity governance : threat modelling with large language modelsNika Jeršič, Muhamed Turkanović, Tina Beranič, 2025, izvirni znanstveni članek Opis: With the increased complexity of applications and systems, threat modelling struggles to keep pace with the evolution of risks. This article addresses this challenge by exploring how large language models (LLMs) can be leveraged to create comprehensive threat models across different risk assessment methodologies. We examine whether a single generic prompt can support frameworks such as LINDDUN, PASTA, and STRIDE, despite their different requirements. Through this comparative analysis, we identify components that enable AI-based assessments, while acknowledging that privacy, regulatory, and dynamic risks require adaptation of the frameworks. Our findings show that a universal guideline is feasible for broad applications, but adaptation is necessary for effective use. Overall, LLM-based threat modelling improves the accessibility, repeatability, and effectiveness of risk analysis and supports stronger and more sustainable practices.
Ključne besede: cybersecurity, large language models, threat modelling, sustainability, resilient infrastructure, SDG 9
Objavljeno v DKUM: 02.12.2025; Ogledov: 0; Prenosov: 7
 Celotno besedilo (875,62 KB) |
2. Large language models for G-code generation in CNC machining: A comparison of ChatGPT-3.5 and ChatGPT-4oKristijan Šket, David Potočnik, Miran Brezočnik, Mirko Ficko, Simon Klančnik, 2025, izvirni znanstveni članek Opis: This research explores the viability of producing ISO G-code for 3-axis machining with OpenAI's Chat Generative Pre-Trained Transformer models, particularly ChatGPT-3.5 and the newer GPT-4o. G-code (RS-274-D, ISO 6983) converts human directives into commands that machines can understand, controlling toolpaths, spindle velocities, and feed rates to produce particular aspects of an object. Previously, G-code was generated either by hand or through the use of computer-aided manufacturing (CAM) software along with machine-specific post-processors, both of which may require considerable time and expense. This research aimed to assess the practicality and effectiveness of specific large language models (LLMs) in generating G-code. The assessment took place in three distinct phases on a sample component that required 3-axis machining. These phases included: (1) the self-generated production of G-code for the sample component, (2) the examination of the independently generated G-code in the CAM application, and (3) the recognition and justification of mistakes in the G-code. The outcomes indicated varying abilities with promising findings. This method could accelerate and possibly enhance manufacturing workflows by decreasing reliance on expensive CAM software and specialized knowledge.
Ključne besede: generative artificial intelligence, intelligent manufacturing, large language models (LLM), ChatGPT, CNC machining, G-code programming
Objavljeno v DKUM: 28.11.2025; Ogledov: 0; Prenosov: 10
Celotno besedilo (4,02 MB)
Gradivo ima več datotek! Več... |
3. Human-led and artificial intelligence-automated critical appraisal of systematic reviews : comparative evaluationLucija Gosak, Gregor Štiglic, Wilson Tam, Dominika Vrbnjak, 2025, izvirni znanstveni članek Opis: Aim To evaluate and compare human-led and artificial intelligence-automated critical appraisal of evidence. Background Critical appraisal is essential in evidence-based practice, yet many nurses lack the skills to perform it. Large language models offer potential support, but their role in critical appraisal remains underexplored. Design We conducted a comparative study to evaluate the performance of five commonly used large language models versus two human reviewers in appraising four systematic reviews on interventions to reduce medication administration errors. Methods We compared large language models and two human reviewers in independently appraising four systematic reviews using the JBI Critical Appraisal Checklist. These models were Perplexity Sonar (Pro), Claude 3.7 Sonnet, Gemini 2.0 Flash, GPT-4.5 and Grok-2. All models received identical full texts and standardized prompts. Responses were analyzed descriptively and agreement was assessed using Cohen’s Kappa. Results Large language models showed full agreement with human reviewers on five of 11 JBI items. Most disagreements occurred in appraising search strategy, inclusion criteria and publication bias. The agreement between human reviewers and large language models ranged from slight to moderate. The highest level of agreement was observed with Claude (κ = 0.732), while the lowest level was observed with Gemini (κ = 0.394). Conclusion Large language models can support aspects of critical appraisal evidence but lack contextual reasoning and methodological insight required for complex judgments. While Claude 3.7 Sonnet aligned most closely with human reviewers, human oversight remains essential. Large language models should serve as adjuncts and not substitutes for evidence-based practice.
Ključne besede: artificial intelligence in healthcare, multimodal large language models, nursing, evidence-based practice
Objavljeno v DKUM: 12.11.2025; Ogledov: 0; Prenosov: 1
 Povezava na datoteko |
4. Detection of malicious software using large language models : master's degree thesisMartina Tivadar, 2025, magistrsko delo Opis: This thesis examines the success rate of large language models (LLM) in detecting macOS malware through Endpoint Security logs. A literature review and 144 experiments with three ChatGPT variants and six prompt types evaluated accuracy, precision, recall, specificity, and F1-score. Results show that prompt wording is crucial: zero-shot and chain-of-thought prompts performed best, while conservative prompts minimized false positives but missed threats. GPT-4o and o1 outperformed o4-mini but showed similar results. Findings suggest LLMs can support, but not replace, traditional detection, with prompt design proving as important as model choice.
Ključne besede: malware, large language models, detection
Objavljeno v DKUM: 03.11.2025; Ogledov: 0; Prenosov: 13
Celotno besedilo (29,72 MB) |
5. Generating test cases for automotive requirement testingusing rag : magistrsko deloMatic Krepek, 2025, magistrsko delo Opis: The automotive industry is increasingly confronted with challenges in managing complex requirements and test cases arising from the integration of advanced electronic systems, software functionalities, and compliance with international standards. Conventional manual validation of requirements is time-consuming, error-prone, and resource-intensive, underscoring the need for more efficient and reliable approaches. This thesis investigates the automation of test case generation through the application of Retrieval-Augmented Generation (RAG) in combination with Large Language Models (LLMs). A complete RAG workflow was implemented in Python, incorporating LangChain, LangGraph, Ollama, and ChromaDB to facilitate indexing, retrieval, and generation. The system was trained and evaluated on datasets comprising automotive requirements and test cases, with experiments examining embedding quality, retrieval strategies, prompt engineering techniques, and generative model parameters. The results demonstrate that RAG is capable of generating high-quality, contextually relevant test cases on consumer-grade hardware, thereby significantly enhancing efficiency, consistency, and productivity relative to manual methods. Furthermore, the findings suggest that RAG-based systems are best positioned as complementary tools that support, rather than replace, human engineers. This research provides a foundation for future work on hybrid retrieval methods, advanced embedding techniques, and the integration of more powerful LLMs into requirement and test case management processes.
Ključne besede: automotive requirements validation, test case generation, large Language Models, Retrieval-Augmented Generation
Objavljeno v DKUM: 01.10.2025; Ogledov: 0; Prenosov: 0
Celotno besedilo (5,84 MB) |
6. The impact of usability and reliability on ChatGPT satisfaction among gen Z and gen YMirjana Pejić Bach, Mirko Palić, Vanja Šimićević, 2025, izvirni znanstveni članek Opis: Background/Purpose: ChatGPT’s rapid diffusion has transformed large-language-model (LLM) technology from a specialist tool into a mainstream companion for study and work. However, empirical evidence on what drives user satisfaction outside medical settings remains scarce. Focusing on future business and management professionals in Croatia, this study examines how perceived ease of use and perceived reliability shape satisfaction with ChatGPT and whether those effects differ between Generation Z (18–25 years) and Generation Y (26–35 years). Methodology: An online survey administered in August 2024 yielded 357 valid responses. The measurement model met rigorous reliability and validity criteria (CFI = 0.96, SRMR = 0.04). Results: Structural-equation modelling showed that, in the pooled sample, ease of use (β = 0.42) and reliability (β = 0.46) jointly explained 72 % of satisfaction. Multi-group analysis revealed a generational split: both predictors were significant for Gen Z. However, only reliability remained significant for Gen Y. Gaussian graphical models corroborated these findings, indicating a densely interconnected attitude network for younger users and a reliability-centred network for older users. Conclusion: The study extends technology-acceptance research to the management domain, underscores the moderating role of generation and illustrates the value of combining SEM with network analytics. Insights inform designers and educators aiming to foster informed, responsible and gratifying engagement with generative AI.
Ključne besede: artificial intelligence, large language models (LLM), marketing, user satisfaction, Croatia, ChatGPT
Objavljeno v DKUM: 04.09.2025; Ogledov: 0; Prenosov: 1
Celotno besedilo (1,06 MB)
Gradivo ima več datotek! Več... |
7. Rethinking English studies through AI : challenges, ethics, and innovationTomaž Onič, Mladen Borovič, David Hazemali, 2025, drugi znanstveni članki Ključne besede: chatbots, artificial intelligence, generative artificial intelligence, large language models, English studies
Objavljeno v DKUM: 27.08.2025; Ogledov: 0; Prenosov: 6
Celotno besedilo (123,95 KB)
Gradivo ima več datotek! Več... |
8. Classifying the information needs of survivors of domestic violence in online health communities using large language models : prediction model development and evaluation studyShaowei Guan, Vivian Hui, Gregor Štiglic, Rose Eva Constantino, Young Ji Lee, Arkers Kwan Ching Wong, 2025, izvirni znanstveni članek Opis: Background: Domestic violence (DV) is a significant public health concern affecting the physical and mental well-being of
numerous women, imposing a substantial health care burden. However, women facing DV often encounter barriers to seeking
in-person help due to stigma, shame, and embarrassment. As a result, many survivors of DV turn to online health communities
as a safe and anonymous space to share their experiences and seek support. Understanding the information needs of survivors of
DV in online health communities through multiclass classification is crucial for providing timely and appropriate support.
Objective: The objective was to develop a fine-tuned large language model (LLM) that can provide fast and accurate predictions
of the information needs of survivors of DV from their online posts, enabling health care professionals to offer timely and
personalized assistance.
Methods: We collected 294 posts from Reddit subcommunities focused on DV shared by women aged ≥18 years who
self-identified as experiencing intimate partner violence. We identified 8 types of information needs: shelters/DV centers/agencies;
legal; childbearing; police; DV report procedure/documentation; safety planning; DV knowledge; and communication. Data
augmentation was applied using GPT-3.5 to expand our dataset to 2216 samples by generating 1922 additional posts that imitated
the existing data. We adopted a progressive training strategy to fine-tune GPT-3.5 for multiclass text classification using 2032
posts. We trained the model on 1 class at a time, monitoring performance closely. When suboptimal results were observed, we
generated additional samples of the misclassified ones to give them more attention. We reserved 184 posts for internal testing
and 74 for external validation. Model performance was evaluated using accuracy, recall, precision, and F1
-score, along with CIs
for each metric.
Results: Using 40 real posts and 144 artificial intelligence–generated posts as the test dataset, our model achieved an F1
-score
of 70.49% (95% CI 60.63%-80.35%) for real posts, outperforming the original GPT-3.5 and GPT-4, fine-tuned Llama 2-7B and
Llama 3-8B, and long short-term memory. On artificial intelligence–generated posts, our model attained an F1
-score of 84.58%
(95% CI 80.38%-88.78%), surpassing all baselines. When tested on an external validation dataset (n=74), the model achieved
an F1
-score of 59.67% (95% CI 51.86%-67.49%), outperforming other models. Statistical analysis revealed that our model significantly outperformed the others in F1
-score (P=.047 for real posts; P<.001 for external validation posts). Furthermore, our
model was faster, taking 19.108 seconds for predictions versus 1150 seconds for manual assessment.
Conclusions: Our fine-tuned LLM can accurately and efficiently extract and identify DV-related information needs through
multiclass classification from online posts. In addition, we used LLM-based data augmentation techniques to overcome the
limitations of a relatively small and imbalanced dataset. By generating timely and accurate predictions, we can empower health
care professionals to provide rapid and suitable assistance to survivors of DV.
Ključne besede: domestic violence, online health communities, large language models, generative artificial intelligence, artificial intelligence
Objavljeno v DKUM: 22.07.2025; Ogledov: 0; Prenosov: 5
Celotno besedilo (780,00 KB)
Gradivo ima več datotek! Več... |
9. Evaluating Proprietary and Open-Weight Large Language Models as Universal Decimal Classification Recommender SystemsMladen Borovič, Eftimije Tomovski, Tom Li Dobnik, Sandi Majninger, 2025, izvirni znanstveni članek Opis: Manual assignment of Universal Decimal Classification (UDC) codes is time-consuming and inconsistent as digital library collections expand. This study evaluates 17 large language models (LLMs) as UDC classification recommender systems, including ChatGPT variants (GPT-3.5, GPT-4o, and o1-mini), Claude models (3-Haiku and 3.5-Haiku), Gemini series (1.0-Pro, 1.5-Flash, and 2.0-Flash), and Llama, Gemma, Mixtral, and DeepSeek architectures. Models were evaluated zero-shot on 900 English and Slovenian academic theses manually classified by professional librarians. Classification prompts utilized the RISEN framework, with evaluation using Levenshtein and Jaro–Winkler similarity, and a novel adjusted hierarchical similarity metric capturing UDC’s faceted structure. Proprietary systems consistently outperformed open-weight alternatives by 5–10% across metrics. GPT-4o achieved the highest hierarchical alignment, while open-weight models showed progressive improvements but remained behind commercial systems. Performance was comparable between languages, demonstrating robust multilingual capabilities. The results indicate that LLM-powered recommender systems can enhance library classification workflows. Future research incorporating fine-tuning and retrieval-augmented approaches may enable fully automated, high-precision UDC assignment systems.
Ključne besede: universal decimal classification, large language models, conversational systems, recommender systems, prompt engineering, zero-shot classification, hierarchical similarity
Objavljeno v DKUM: 21.07.2025; Ogledov: 0; Prenosov: 11
Celotno besedilo (447,50 KB)
Gradivo ima več datotek! Več... |
10. Evaluating chatbot assistance in historical document analysisDavid Hazemali, Janez Osojnik, Tomaž Onič, Tadej Todorović, Mladen Borovič, 2024, izvirni znanstveni članek Opis: The article explores the potential of PDFGear Copilot, a chatbot-based PDF editing tool, in assisting with the analysis of historical documents. We evaluated the chatbot's performance on a document relating to the Slovenian War of Independence. We included 25 factual and 5 interpretative questions to address its formal characteristics and content details, assess its capacity for in-depth interpretation and contextualized critical analysis, and evaluate the chatbot’s language use and robustness. The chatbot exhibited some ability to answer factual questions, even though its performance varied. It demonstrated proficiency in navigating document structure, named entity recognition, and extracting basic document information. However, performance declined significantly in tasks such as document type identification, content details, and tasks requiring deeper text analysis. For interpretative questions, the chatbot's performance was notably inadequate, failing to link cause-and-effect relationships and provide the depth and nuance required for historical inquiries.
Ključne besede: chatbots, historical document analyses, Yugoslavia, wars, generative, artificial intelligence, large language models
Objavljeno v DKUM: 18.07.2025; Ogledov: 0; Prenosov: 10
Celotno besedilo (1,20 MB)
Gradivo ima več datotek! Več... |
