Papers from conferences and scientific journals will be published in this section. ( Access password: StopETF! )

• D.Drašković et al., "Otkrivanje govora mržnje vođeno veštačkom inteligencijom" (in Serbian; english "AI-driven hate speech detection")

  "Otkrivanje govora mržnje vođeno veštačkom inteligencijom", Zbornik radova 30. IKT konferencije "YU INFO 2024", Informaciono društvo Srbije, pp. 182-186, Kopaonik, Serbia, March 2024 [ISBN: 978-86-85525-31-5]
  Link: Read paper |
  Link: Proceedings of the conference | Download PDF
  Abstract: Hate speech is anything that offends an individual, population or phenomenon on any basis, which can be sexual orientation, belonging to a religion, nationality, race, belonging to a certain group, expressed views or other. Detecting and reducing hate speech is critical to the safety and well-being of individuals, as such speech can otherwise lead to harm and tragedy in the real world. Traditional methods of manually monitoring content on the Internet are time-consuming, expensive, and inefficient when dealing with huge amounts of user content. Therefore, there is a need for the development of automated tools that can effectively detect or prevent the publication of hate speech, and which constitute the primary objectives of this research. This research represents a significant milestone as the first initiative for the development of a hate speech detection model designed for the Serbian language. In this paper, an analysis of existing techniques that detect and classify hate speech in other languages is given, and a model proposal is also given, which will be developed for the detection and classification of hate speech in the Serbian language. The research is conducted within the national project “Software for Text Offences Prevention in Serbian: Al-driven Hate Speech Detection”, financed by the Science Fund of the Republic of Serbia in the period 2024-2025.

• M.Dodović et al., "Enhancing Sentiment Analysis in Product Reviews: Fine-Tuning BERT for Class Imbalance and Optimal Sequence Representation"

  M.Dodović, M.Ogrizović, D.Miladinović, D.Drašković, "Enhancing Sentiment Analysis in Product Reviews: Fine-Tuning BERT for Class Imbalance and Optimal Sequence Representation", Springer's Lecture Notes in Networks and Systems, with title Disruptive Information Technologies for a Smart Society, vol 860, Springer, Cham, Kopaonik, March 2024 [DOI: https://doi.org/10.1007/978-3-031-71419-1_30]
  Link: View paper in Proceedings | Download PDF
  Abstract: Sentiment analysis, a pivotal aspect of NLP (Natural Language Processing), of-fers profound insights into the public sentiment from vast swathes of unstruc-tured textual data. This paper presents an empirical investigation into the applica-bility and effectiveness of the BERT (Bidirectional Encoder Representations from Transformers) algorithm for sentiment analysis, particularly focused on product reviews. The research delves into the nuances of consumer language expressions and evaluates the capacity of BERT to accurately classify sentiment in a large-scale dataset of food product reviews. The results achieved through this research are significant, with the fine-tuned BERT model demonstrating high accuracies, indicating its robustness and suitability for the sentiment classification task. In addressing the challenges posed by the varying lengths of consumer reviews, this study offers a methodological analysis for selecting the optimal max_seq_length parameter within BERT’s framework. A critical balance is achieved between computational efficiency and the comprehensive inclusion of informative content within the reviews. Furthermore, the paper confronts the prevalent issue of class imbalance in sentiment analysis datasets by employing a weighted loss function during the training of BERT. This technique ensures equi-table representation and consideration of all sentiment classes, enhancing the model's accuracy and fairness.

• J.Cincović et al., "Applied Artificial Intelligence in Detection Hate Speech"

  J.Cincović, U.Radenković, M.Mićović, A.Milaković, V.Jocović, D.Drašković, "Applied Artificial Intelligence in Detection Hate Speech", in publication Applied Artificial Intelligence 4: Medicine, Biology, Chemistry, Financial, Games, Engineering AAI 2024 - Lecture Notes in Networks and Systems, vol 1446, Springer, Cham, Kragujevac, May 2024 [DOI: https://doi.org/10.1007/978-3-031-99201-8_11]
  Link: View paper in Proceedings | Download PDF
  Abstract: The problem this research addresses is detecting hate speech (HS) in texts in Serbian on the Internet. Detecting and reducing HS is crucial for the safety and well-being of individuals, as it can otherwise lead to real-world harm and tragedies. Traditional methods of manually monitoring online content are time-consuming, costly, and ineffective in dealing with the vast amount of user-generated content. Therefore, there is a need for automated tools that can efficiently detect and prevent HS, which constitutes the primary goal of this research. In this paper, we describe the methods of data collection that we intend to implement, present a short analysis of the existing solutions, and provide the conceptual solution of the model.

• D.Drašković et al., "Data analysis techniques and detection of propaganda in Serbian online media in 2023"

  D.Drašković, M.Ogrizović, M.Dodović, M.Obradović, "Data analysis techniques and detection of propaganda in Serbian online media in 2023", in Applied Artificial Intelligence 4: Medicine, Biology, Chemistry, Financial, Games, Engineering, AAI 2024 - Lecture Notes in Networks and Systems, vol 1446, Springer, Cham, Kragujevac, May 2024 [DOI: https://doi.org/10.1007/978-3-031-99201-8_12]
  Link: View paper in Proceedings | Download PDF
  Abstract: Propaganda is a technique by which people manipulate information to direct the behavior of an individual or group, with the aim of provoking a desired reaction. Propaganda is strongly expressed in the spread of religion, political parties, and autocratic governments. In this paper, the most common propaganda techniques are presented, which propaganda implementers can use in shaping and realizing their messages. A certain content of short texts published on the portals of media companies in Serbia was collected. Then, using machine learning techniques, a model was developed, which was trained to classify the text as propaganda or not propaganda. This research included texts that were published on the portals RTS, Politika, B92, Blic, N1, and Danas in 2023. The texts were collected in real-time, because some media portals do not have an archive.

• L.Hrvačević et al., "Development of a web system with an automated question generator based on large language models"

  Hrvačević Luka, Cincović Jelica, Milaković Adrian, Jocović Vladimir, Matvejev Valerijan, Drašković Dražen, "Development of a web system with an automated question generator based on large language models," 2024 11th International Conference on Electrical, Electronic and Computing Engineering (IcETRAN), Niš, Serbia, 2024, pp. 1-6 [DOI: 10.1109/IcETRAN62308.2024.10645106].
  Link: View paper in Proceedings | Download PDF
  Abstract: In this paper, we describe the possibility of using large language models (LLMs) in education through the development of a modern web system. A web-based platform employing LLMs to produce questions and answers spanning diverse domains, drawing upon Serbian Wikipedia as its primary knowledge repository. Leveraging LLMs for question-and-answer generation represents a novel methodology in quiz development. Additionally, incorporating Wikipedia enhances the accuracy and pertinence of the generated questions, thereby enriching the overall user experience.

• J. Tufegdžić et al., "Application of WebAssembly technology in high-performance web applications"

  Tufegdžić Janko, Dodović Matija, Ogrizović Mihajlo, Babić Nikola, Đukić Jovan, Drašković Dražen, "Application of WebAssembly technology in high-performance web applications," 2024 11th International Conference on Electrical, Electronic and Computing Engineering (IcETRAN), Niš, Serbia, 2024, pp. 1-6 [DOI: 10.1109/IcETRAN62308.2024.10645198].
  Link: View paper in Proceedings | Download PDF
  Abstract: This research paper explores the efficacy of WebAssembly technology in enhancing the performance of web applications, with a focus on graphic-intensive domains such as video games, simulations, and image processing. WebAssembly, a binary instruction format designed for browsers, offers performance comparable to desktop applications by executing code within a secure, low-level virtual machine environment. The study further examines the integration of WebAssembly into various programming languages such as C/C++, C#, and Rust, highlighting its cross-platform capabilities and efficient memory management. Through the development of a website that hosts retro-themed video games, the paper assesses performance across different platforms, demonstrating that while unrestricted desktop platforms show higher speeds, regulated environments exhibit negligible performance discrepancies. The findings suggest that while WebAssembly substantially boosts web application performance, opportunities for further enhancements in areas such as garbage collection, debugging, and JavaScript integration remain. This research underscores WebAssembly's potential to revolutionize web application development, providing a robust framework for future advancements in complex web-based applications.

• M.Ogrizović, D.Drašković, D.Bojić, "Quality assurance strategies for machine learning applications in big data analytics: an overview"

  Ogrizović, M., Drašković, D., Bojić, D., "Quality assurance strategies for machine learning applications in big data analytics: an overview," Journal of Big Data 11, 156 (2024) [DOI: https://doi.org/10.1186/s40537-024-01028-y]
  Link: View paper | Download PDF
  Abstract: Machine learning (ML) models have gained significant attention in a variety of applications, from computer vision to natural language processing, and are almost always based on big data. There are a growing number of applications and products with built-in machine learning models, and this is the area where software engineering, artificial intelligence and data science meet. The requirement for a system to operate in a real-world environment poses many challenges, such as how to design for wrong predictions the model may make; How to assure safety and security despite possible mistakes; which qualities matter beyond a model’s prediction accuracy; How can we identify and measure important quality requirements, including learning and inference latency, scalability, explainability, fairness, privacy, robustness, and safety. It has become crucial to test thoroughly these models to assess their capabilities and potential errors. Existing software testing methods have been adapted and refined to discover faults in machine learning and deep learning models. This paper covers a taxonomy, a methodologically uniform presentation of all presented solutions to the aforementioned issues, as well as conclusions about possible future development trends. The main contributions of this paper are a classification that closely follows the structure of the ML-pipeline, a precisely defined role of each team member within that pipeline, an overview of trends and challenges in the combination of ML and big data analytics, with uses in the domains of industry and education.

• A.Milaković et al., "Detecting ugly and derogatory words in Serbian language using a web browser extension"

  A. Milaković, V. Jocović, J. Cincović, M. Mićović, U. Radenković, D.Drašković, "Detecting ugly and derogatory words in Serbian language using a web browser extension", 2024 32nd Telecommunications Forum (TELFOR), Belgrade, Serbia, November 2024, pp. 1-4 [DOI: https://doi.org/10.1109/TELFOR63250.2024.10819059]
  Link: View paper in Proceedings | Download PDF
  Abstract: In natural language processing, data acquisition and preprocessing techniques are significant for experiments involving training models on cleaned data. This paper describes the formation of a dataset of ugly and derogatory words in the Serbian language and the development of a web extension to analyze texts in which such words appear and then their censoring. Developing such an add-on is essential in creating a safer and healthier digital environment for young users who are just developing their Internet habits and adults who want to avoid unwanted content. It is crucial to point out that there are few software tools for natural languages with limited resources, such as the Serbian language. This software could analyze the frequency of certain foul words on different sites, which would contribute to a better understanding of the impact of such language on users.

• D.Drašković et al., "Practical teaching in the field of software engineering"

  D.Drašković, T.Šekularac-Obradović, M.Ogrizović, L.Hrvačević, T.Radaljac, D.Bojić, "Practical teaching in the field of software engineering", 20th China - Europe International Symposium on Software Engineering Education (CEISEE 2024), Belgrade, Serbia, 2024. (organised by Harbin Institute of Technology and University of Belgrade) [ISBN: 978-86-85525-34-6]
  Link: View Conference Program | Download paper (PDF)
  Abstract: Considering the estimates of the growing demand for software engineers in Serbia and the world, the University of Belgrade - School of Electrical Engineering founded a modern study program in Software Engineering in 2004. In this study program, the segment of the program related to classical electrical engineering has been significantly reduced, but specialized courses are taken from the second semester. A perfect side of this study program is the large number of optional subjects, as well as the great attention focused on practical teaching. Students in this study program are expected to complete homeworks starting from the first year, and in the later years of study, they are working on are larger independent or team projects. This paper describes practical teaching in the subjects of the basics of software engineering, such as principles of software engineering, software design, software testing, and software project management. Ways of working with students through homework and project assignments, laboratory exercises, methods of knowledge transfer, scoring systems, and software tools included in the teaching process are highlighted.

• D.Drašković et al., "Visual simulators in the teaching of artificial intelligence"

  D.Drašković, A.Milaković, V.Jocović, B.Nikolić, "Visual simulators in the teaching of artificial intelligence", 20th China - Europe International Symposium on Software Engineering Education (CEISEE 2024), Belgrade, Serbia, 2024. (organised by Harbin Institute of Technology and University of Belgrade) [ISBN: 978-86-85525-34-6]
  Link: View Conference Program | Download paper (PDF)
  Abstract: Teaching machine learning (ML) concepts to students who have no programming experience can be very challenging. In 2021, a visual simulator was introduced in the course Intelligent Systems at the Faculty of Electrical Engineering of the University of Belgrade. This simulator allows students to learn key machine learning concepts without programming, from data selection and preprocessing, hyperparameter selection, and model training. This paper describes the importance of using such simulators in teaching. The paper provides an insight into the results of student success from the previous 6 academic years, which show an increase in student performance through higher passing rates, better grades and a greater number of students achieving the highest grades. This analysis suggests that simulators have an important role in the learning process of complex subjects.

• D.Drašković et al., "Prikupljanje podataka i označavanje u procesu formiranja skupova podataka govora mržnje na srpskom jeziku" (in Serbian; in English: "Data collection and annotation in the process of forming datasets hate speech in the Serbian language")

  D.Drašković, U.Radenković, M.Mićović, J.Cincović, A.Milaković, V.Jocović, "Prikupljanje podataka i označavanje u procesu formiranja skupova podataka govora mržnje na srpskom jeziku", Zbornik radova 31. IKT konferencije "YU INFO 2025", Informaciono društvo Srbije, pp. 93-98, Kopaonik, Serbia, March 2025 [ISBN: 978-86-85525-33-9]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: Hate speech represents a serious social and technological challenge in the contemporary digital environment, particularly due to its widespread presence on social networks, online forums, and media platforms. The rapid growth of user-generated content, combined with anonymity and the lack of effective moderation, has significantly contributed to the proliferation of discriminatory, offensive, and violent discourse. In this context, the development of high-quality datasets is a crucial prerequisite for building reliable automated systems for hate speech detection, especially for languages with limited digital resources, such as Serbian. This paper presents a comprehensive overview of the data collection and annotation process used in the formation of a hate speech dataset in the Serbian language. The dataset was created within a national research project aimed at preventing textual offenses through artificial intelligence methods. Data were collected from multiple sources, including social networks, online news portals, and public discussion platforms, using both automated techniques (APIs, web crawlers, and scrapers) and manual procedures. The resulting corpus consists of 4,300 short text samples. A three-class annotation scheme was adopted, distinguishing between non-hateful content, offensive language, and hate speech, in accordance with European Union guidelines and relevant legal and ethical frameworks. In addition, hate speech instances were further categorized based on different forms of discrimination, such as national, ethnic, religious, gender-based, and political hatred. The paper also provides statistical and qualitative analyses of the dataset, including class distribution, text length characteristics, and the most frequent terms associated with hate speech. The presented dataset represents a valuable resource for future research in natural language processing and artificial intelligence, enabling the development and evaluation of automated hate speech detection models for the Serbian language.

• A. Mandić et al., "Primena veštačke inteligencije u analizi i obradi govora kroz transkripciju,verifikaciju i evaluaciju izjava" (in Serbian; in English: "Application of artificial intelligence in speech analysis and processing through transcription, verification and evaluation of statements")

  Ana Mandić, Marko Jelović, Ana Bulatović, Filip Nikolić, Hana Mijatović, Nikola Vučićević, Ana Stanić, Natalija Bogdanović, Luka Lazović, Anja Mihajlov, Lana Popović, Adrian Milaković, Vladimir Jocović, Dražen Drašković, "Primena veštačke inteligencije u analizi i obradi govora kroz transkripciju,verifikaciju i evaluaciju izjava", Zbornik radova 31. IKT konferencije "YU INFO 2025", Informaciono društvo Srbije, pp. 99-103, Kopaonik, Serbia, March 2025 [ISBN: 978-86-85525-33-9]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: In a world where information spreads at an exceptional speed, it is often challenging to distinguish accurate from inaccurate data obtained from the media. Manual analysis of public statements requires a significant amount of time and resources, making it impractical for processing large volumes of information. To address this challenge, a system has been developed that combines the power of large language models with web search, enabling automated fact-checking and the delivery of reliable results. Such technology has a wide range of applications—from media organizations and research teams to anyone seeking trustworthy information and a better understanding of the content presented to them.

• V.Matvejev, D.Drašković, "Predviđanje uspešnosti memorizacije slike pomoću modela mašinskog učenja" (in Serbian; in English: "Predicting Image Memorability Using Machine Learning Models")

  V.Matvejev, D.Drašković, "Predviđanje uspešnosti memorizacije slike pomoću modela mašinskog učenja", Zbornik radova 31. IKT konferencije "YU INFO 2025", Informaciono društvo Srbije, pp. 62-65, Kopaonik, Serbia, March 2025 [ISBN: 978-86-85525-33-9]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: Visual attention, i.e., the automatic selection of the most important information within a visual stimulus, is a topic in computer vision that has attracted significant scientific interest. This research focuses on studying the complex relationship between implicit, covert attention and explicit, overt attention, as well as human memory, through a memorability experiment conducted on static images. Thirty images used in the experiment were selected from the FIGRIM image dataset.

• M.Dodović et al., "From Documents to Answers: A University-Focused DBQA System with LLMs"

  M.Dodović, J.Tufegdžić, L.Hrvačević, D.Drašković, "Enhancing Sentiment Analysis in Product Reviews: Fine-Tuning BERT for Class Imbalance and Optimal Sequence Representation", Springer's Lecture Notes in Networks and Systems, with title Transformative Technologies Shaping a Smarter Society ICIST 2025, vol. 1621, pp. 333-343, Springer, Cham, Kopaonik, March 2024 [DOI: https://doi.org/10.1007/978-3-032-04890-5_25]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: This paper presents a document-based question-answering (DBQA) system utilizing a large language model (LLM) to assist students in retrieving relevant information from official university documents. The system effectively processes machine-readable documents by leveraging pre-trained multilingual embeddings and a transformer-based QA model while supporting both Serbian and English queries. The experimental setup focused on extracting relevant context using cosine similarity and answering queries using a fine-tuned QA pipeline, achieving an average accuracy of 84.16%. Evaluation results highlight the system’s strengths in retrieving factual answers while identifying challenges related to answer format variations and partial extractions. This research demonstrates the potential of LLM-based DBQA systems in academic and institutional environments, improving information accessibility for students.

• L.Hrvačević et al., "A Comprehensive Review of Large Language Models for Document-Based Question Answering"

  L.Hrvačević, M.Dodović, J.Tufegdžić, D.Drašković, "A Comprehensive Review of Large Language Models for Document-Based Question Answering", Springer's Lecture Notes in Networks and Systems, with title Transformative Technologies Shaping a Smarter Society ICIST 2025, vol. 1621, pp. 289-301, Springer, Cham, Kopaonik, March 2024 [DOI: https://doi.org/10.1007/978-3-032-04890-5_22]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: This review paper provides a comprehensive survey of Large Language Models (LLMs) applied to Document-Based Question Answering (DBQA). The rapid advancements in LLMs have transformed DBQA systems by enhancing their capabilities in complex reasoning, multi-hop question answering, and contextual understanding. We explore the historical evolution of QA systems before the introduction of LLMs, followed by an in-depth analysis of datasets, benchmarks, and evaluation metrics that underpin DBQA performance. A key contribution of this work is the categorization of LLMs that can perform DBQA based on linguistic focus, distinguishing between monolingual models tailored for specific languages and multilingual models designed to operate across diverse linguistic contexts. Furthermore, we examine notable LLM architectures, their domain-specific adaptations in areas like healthcare and law, and their performance in handling both structured and unstructured data.

• M.Vučinić et al., "Expense tracker using LLM: Mobile application with scanning receipts issued in Serbia"

  M.Vučinić, U.Radenković, M.Mićović, V.Jocović, "Expense tracker using LLM: Mobile application with scanning receipts issued in Serbia", Springer's Lecture Notes in Networks and Systems, with title Transformative Technologies Shaping a Smarter Society ICIST 2025, vol. 1621, pp. 517-532, Springer, Cham, Kopaonik, March 2024 [DOI: https://doi.org/10.1007/978-3-032-04890-5_38]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: With the increase in available products and services, unplanned spending has become an indispensable part of everyday life. Planning finances and savings for the future has become a struggle for many. Numerous methods that can be used to track expenses, both modern and traditional, are available. Mobile applications are becoming increasingly popular for tracking and planning personal finances, over traditional approaches like paper and spreadsheets, especially among the younger generation. This paper proposes the implementation of the expense tracking application, providing functionalities that will accelerate input of expenses, such as adding transactions with receipt scanning and automatic categorization. Users can easily add transactions, through an intuitive user interface, by scanning the QR code on the receipt (issued in the Republic of Serbia) or manually. The bill data and items are fetched from the Tax Administration office website. Expenses are automatically categorized and summarized based on bill items using the Large Language Models (LLMs) with minimal user engagement. Classifying costs based on bill items leads to greater precision compared to classifying based on location which is used by most other applications. The charts show the amount of money spent across categories on a monthly and annual level. It is possible to set a limit for each month and observe whether the limit is exceeded over time to achieve more sustainable spending.

• D.Drašković, "Integration of AI tools into an AI-driven software system to make learning programming easier"

  D.Drašković, "Integration of AI tools into an AI-driven software system to make learning programming easier", 2025 IEEE Global Engineering Education Conference (EDUCON), London, United Kingdom, 2025, pp. 1-6 [DOI: https://doi.org/10.1109/EDUCON62633.2025.11016313]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: Today, a person can be considered fully digitally literate if they know how to use and integrate ready-made artificial intelligence (AI) tools. The use of AI tools is becoming increasingly common in students' learning processes. However, learning programming can be challenging and exhausting, especially for younger learners. In this research, a software system was developed to integrate multiple AI tools to facilitate the learning process. The system includes tools for speech and text processing, program code generation, code testing, and result verification. Through a user-friendly software interface, users can define a problem or programming task using speech. The software then converts the speech into text using the Whisper AI API, which is subsequently processed by the GPT-3.5 Turbo and Claude AI APIs to generate program code. Once the program code is generated, it undergoes a series of tests, including parallel testing on the LeetCode platform. Users then compare the obtained results and manually complete a survey evaluating both external tools. One key research requirement was for the software system to accept input data in Serbian, a language with limited resources and complex grammatical rules. This made it difficult to find a suitable AI tool for accurate speech-to-text transformation. The system was tested with speech in both English and Serbian but supports many additional languages thanks to the powerful Whisper AI API. The implemented system is modular and easily extensible with new APIs, making it applicable to other areas of education beyond programming.

• V.Milutinović et al., "Applied AI, Teamwork, and Learning in Student Hackathons: Case Study from Serbia"

  V.Milutinović, J.Cincović, V.Jocović, D.Drašković, "Applied AI, Teamwork, and Learning in Student Hackathons: Case Study from Serbia", 2025 12th International Conference on Electrical, Electronic and Computing Engineering (IcETRAN), Cacak, Serbia, 2025, pp. 1-6 [DOI: https://doi.org/10.1109/IcETRAN66854.2025.11114198]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: Student hackathons have emerged as a dynamic educational mechanism for fostering creativity and accelerating the adoption of new technologies among Computer Science and Software Engineering students. First introduced in university settings in the early 2000s, hackathons have evolved into highly engaging events where students compete to develop innovative software or hardware solutions within a limited time frame. The competitive spirit inherent to hackathons resonates strongly with today's engineering students, motivating them to push boundaries, solve real-world problems, and collaborate effectively under pressure. Participation in hackathons enhances students' technical skills, promotes teamwork, and deepens their understanding of modern tools, frameworks, and platforms not always covered in formal curricula. Over the past five years, hackathons have gained remarkable popularity in Serbia, especially at state faculties and through active student organizations. Many of these events are organized in partnership with globally recognized ICT companies, providing students with valuable exposure to industry standards and expectations. At the University of Belgrade, the students' programmers' club, within the Google Developer Group on Campus program, has successfully hosted numerous hackathons that have become key platforms for skill development and innovation. These events not only contribute to academic growth but also significantly enhance students' employability and entrepreneurial thinking. Through hands-on experience and interaction with industry mentors, participants often gain clarity about their professional interests and career paths. Hackathons thus serve as bridges between university education and real-world software engineering practice, preparing students for both modern studies and the tech industry.

• S.Stanković et al., "Comparative Analysis of Machine Learning and Federated Learning in Healthcare"

  S.Stanković, P.Vuletić, D.Drašković, "Comparative Analysis of Machine Learning and Federated Learning in Healthcare", 2025 12th International Conference on Electrical, Electronic and Computing Engineering (IcETRAN), Cacak, Serbia, 2025, pp. 1-6 [DOI: https://doi.org/10.1109/IcETRAN66854.2025.11114237]
  Link: View Conference Proceedings | Download paper (PDF)
  Abstract: With a rapid development of artificial intelligence in healthcare, machine learning and federated learning has emerged as critical technologies for improving decision-making, diagnostics and personalized treatments. This paper presents a comprehensive comparison between traditional machine learning and newer federated approach, focusing on their applications in healthcare. Even though machine learning provides great efficiency in resolving complex tasks in healthcare, it faces great challenges such as data privacy, security and communication cost. In contrast, federated learning offers a decentralized approach which resolves these limitations and allows healthcare institutions to collaboratively train models across multiple devices and locations without sharing sensitive patient data. The experiment conducted in this study utilized a medical dataset to predict whether an individual is an alcohol consumer or not. A range of metrics, including loss, accuracy, precision, recall, AUC, and F1 score, were employed to evaluate and compare the performances of both approaches. By leveraging this technique, federated learning addresses privacy concerns, while maintaining high model performances across distributed dataset. Experiments conducted on healthcare datasets highlight the strengths, challenges and limitations of both machine and federated learning in practical, real-world scenarios.

• D.Bojić, D.Drašković, "Modernizing 90s Era Software to a New Language and Environment Using LLMs - An Empirical Investigation"

  D.Bojić, D.Drašković, "Modernizing 90s Era Software to a New Language and Environment Using LLMs - An Empirical Investigation", INTERNATIONAL JOURNAL OF SOFTWARE ENGINEERING AND KNOWLEDGE ENGINEERING, Vol. 35, No. 8, pp. 1099 - 1119, May, 2025 [DOI: https://doi.org/10.1142/S021819402550024X]
  Link: View Paper in Journal | Download paper (PDF)
  Abstract: Legacy software, particularly from the 1990s, often becomes obsolete due to aging hardware and outdated software environments. Traditionally, software modernization required extensive manual effort, involving reverse engineering, code rewriting, and re-architecting. However, advancements in large language models (LLMs) have introduced new possibilities for automating software translation and modernization. This paper explores the feasibility of using LLMs for modernizing 90s-era Windows applications, specifically migrating legacy C and C++ code to Python. Our methodology includes decompilation, source code analysis, automated translation using ChatGPT, and user interface reconstruction. We empirically evaluate three software projects by analyzing LLM-based translation accuracy across different code structures, including algorithmic logic, file handling, and graphical interfaces. Results indicate that while LLMs achieve high translation accuracy (~88%) for structured code, challenges persist in handling decompiled code and user interface generation. The study provides insights into the effectiveness and limitations of LLMs in real-world software renovation, offering guidelines for leveraging machine learning in legacy system modernization. These findings contribute to both academic research and practical applications, suggesting a pathway for cost-effective and scalable legacy software migration.

• D.Drašković, S.Milanović, "Aspect-based sentiment analysis of user-generated content from a microblogging platform"

  Drašković, D., Milanović, S., "Aspect-based sentiment analysis of user-generated content from a microblogging platform," Journal of Big Data 12, 186 (2025) [DOI: https://doi.org/10.1186/s40537-025-01244-0]
  Link: View paper | Download PDF
  Abstract: This research presents a practical application of advanced natural language processing techniques to understand people’s feelings during the global Covid-19 pandemic, using a set of big data of over 547 thousand tweets. Companies often use sentiment analysis to process comments, product usage, social media posts, and more, in order to better understand user needs and preferences. In this research paper, aspect-based sentiment analysis is applied as one of the most recent and advanced subtypes of sentiment analysis. Aspect-based sentiment analysis is a modern natural language processing technique that does not perform sentiment detection at the level of the entire input text but individually over all the aspects detected in it. Therefore, this technique enables the precision of analyzing the user data set and making concrete conclusions about people’s feelings. This research aims to develop a software infrastructure for further work in natural language processing using the aspect-based sentiment analysis technique. The proposed process flow and data handling methods, as defined in this research, are designed to be easily adaptable to other data sets with minimal modifications.

• U.Radenković et al., "Can Large Language Models Replace Human Annotators? A Study on Hate Speech Detection in Serbian"

  U. Radenković, J. Cincović, A. Milaković, M. Mićović, M. Dodović, V. Jocović, D. Drašković, "Can Large Language Models Replace Human Annotators? A Study on Hate Speech Detection in Serbian," Proceedings of the 34th International Electrotechnical and Computer Science Conference ERK 2025, pp. 409 - 414, IEEE Slovenia Section with Faculty of Electrical Engineering University of Ljubljana, Portorož, Slovenia, Sep, 2025 [ISSN/ISBN: 2591-0442 (online)]
  Link: Download paper (PDF)
  Abstract: This study focuses on the comparison between human annotation and large language model (LLM)-based annotation of hate speech in Serbian. The dataset consists of short user-generated texts collected from online media portals and social networks. Manual annotation was performed by an expert annotator following detailed labeling guidelines, serving as the reference standard. Automated annotation was conducted using several state-of-the-art LLMs, including GPT-5, Gemini 2.5 Pro, Claude 4, and DeepSeek-R1. The results highlight a strong alignment between human annotation and LLM-based annotation, with GPT-5 achieving the closest agreement with the human annotator. Furthermore, comparative analysis among the top-performing LLMs reveals that, with appropriate fine-tuning, automated annotation can be considered a reliable complement, or in certain contexts, an alternative to manual labeling.

• J.Tufegdžić et al., "LLM-Driven Hate Speech Detection in News Content Extracted from URLs"

  J. Tufegdžić, L. Hrvačević, M. Dodović, N. Stanković, D. Drašković, "LLM-Driven Hate Speech Detection in News Content Extracted from URLs," Proceedings of the 34th International Electrotechnical and Computer Science Conference ERK 2025, pp. 415 - 420, IEEE Slovenia Section with Faculty of Electrical Engineering University of Ljubljana, Portorož, Slovenia, Sep, 2025 [ISSN/ISBN: 2591-0442 (online)]
  Link: Download paper (PDF)
  Abstract: This paper presents a system for detecting hate speech in Serbian language in online news articles using large language models (LLMs). While most prior work focuses on social media platforms, news content poses unique challenges due to more subtle, implicit, and editorially framed expressions of hate speech. Our approach integrates a web scraper that extracts clean text from article URLs and a prompt-based LLM classifier that highlights hateful segments without altering the original structure. The research analyzes content in the Serbian language, which is one of the languages with limited resources, but the methodology can be adapted and applied to other languages. To evaluate the system, we tested it on live articles from online Serbian media known for publishing content with hate speech, based on previously annotated datasets from related 2023 studies. We evaluate four state-of-the-art LLMs: LLaMA 3, Mistral, Phi-3, and Qwen3, on a curated dataset of Serbian news articles with human-labeled annotations. Results show that LLaMA 3 most closely aligns with human judgments, while Qwen3 tends to over-tag, and Mistral and Phi-3 often under-tag hate speech. The entire system, including the web interface, is available as open source. Our findings highlight the potential of LLMs for reliable hate speech detection in formal, long-form content and demonstrate the importance of model choice and prompt design for balanced moderation.

• M.Dodović et al., "From Toxic to Neutral: Automatic Hate Speech Rewriting Using LLMs"

  M. Dodović, N. Stanković, J. Tufegdžić, L. Hrvačević, D. Drašković, "From Toxic to Neutral: Automatic Hate Speech Rewriting Using LLMs," Proceedings of the 34th International Electrotechnical and Computer Science Conference ERK 2025, pp. 421 - 426, IEEE Slovenia Section with Faculty of Electrical Engineering University of Ljubljana, Portorož, Slovenia, Sep, 2025 [ISSN/ISBN: 2591-0442 (online)]
  Link: Download paper (PDF)
  Abstract: This paper presents a system for rewriting hate speech in the Serbian language using large language models (LLMs) to produce culturally appropriate and neutral reformulations. Our approach centers around content purification, which includes removal of hateful, offensive, or discriminatory elements from a sentence while preserving its original intent when possible. The research analyzes content in the Serbian language, which is one of the languages with limited resources, but the methodology can be adapted and applied to other languages as well. The core of the system is a carefully designed prompt, instructing the model to either rewrite the sentence with improved tone or replace it entirely with a neutral message if there is no valuable content. We evaluate four state-of-the-art LLMs: LLaMA 3, Mistral, Phi-3, and Qwen3, using a curated dataset of hate speech examples in Serbian, extracted from comments left on Serbian news articles. The models are run locally using the Ollama framework, ensuring full data privacy and control. Results indicate that Mistral and Qwen3 offer the most consistent and context-aware rewrites, while other models show varying sensitivity to different types of hate speech. Our findings emphasize the effectiveness of prompt-based LLM interaction for hate speech moderation and highlight the importance of prompt design, and model selection in sensitive language-processing tasks.

• V.Matvejev et al., "Web Application for Predicting Image Memorability Using a Machine Learning Model"

  V.Matvejev, D.Drašković, B.Nikolić, D.Bojić, "Web Application for Predicting Image Memorability Using a Machine Learning Model", 2025 33rd Telecommunications Forum (TELFOR), Belgrade, Serbia, 2025, pp. 1-4 [DOI: https://doi.org/10.1109/TELFOR67910.2025.11314318]
  Link: View paper | Download paper (PDF)
  Abstract: Visual attention i.e. the automatic selection of the most important information within a visual stimulus - is a topic within computer vision that attracts considerable scholarly interest. In this study, the focus was on examining the complex relationship between covert (implicit, hidden) and overt (explicit, observable) attention and human memory through a memorability experiment on static images. Thirty images for the experiment were selected from the FIGRIM dataset. The collected eye-tracking data were first used to compute fixation maps and visual saliency maps, and then the IOVC (Inter-Observer Visual Congruency) metric; together with the percentage of successful image memorization (the memorability score), these served as the key parameters of the study. A significant correlation between the IOVC score and the memorability score indicated the possibility of automating the prediction of whether viewing an image would lead to successful memorization based solely on its fixation map. Inspired by this insight, we first trained a machine learning model on the collected viewing data to infer, from a fixation map, which of the 30 images had been viewed. The goal was to train a classifier capable of distinguishing images using only their viewing (eye-tracking) data. The trained model achieved close to 90% accuracy in determining whether an arbitrary fixation map would lead to successful memorization of the given image. A web application that could be used for commercial use for memorability prediction was developed.

• D.Drašković et al., "Using generative AI in hate speech detection for primary and secondary school students"

  D.Drašković, J.Cincović, U.Radenković, V.Jocović, M.Mićović, A.Milaković, "Using generative AI in hate speech detection for primary and secondary school students", 2025 IEEE International Conference on Teaching, Assessment, and Learning for Engineering (TALE), pp. 1 - 7, IEEE, Macao, Dec, 2025 [DOI: https://ieeexplore.ieee.org/document/11346657]
  Link: View paper | Download paper (PDF)
  Abstract: Hate speech among children in primary and secondary schools represents a widespread and concerning phenomenon. It is unrealistic to expect young individuals to refrain from discriminatory behavior if they are regularly exposed to hateful rhetoric in their environment. This research focuses on the design and development of a web-based application for the detection and classification of hate speech in the Serbian language, with the broader goal of supporting efforts to reduce hate speech among school-aged children. Serbian, as a South Slavic language with limited natural language processing resources, poses additional challenges for computational analysis compared to globally dominant languages due to the scarcity of publicly available datasets. The implemented software tool leverages generative artificial intelligence, specifically state-of-the-art large language models including GPT-4o, Claude 4, Gemini 2.5-Pro, Llama 3, and DeepSeek-R1. The tool is intended for use in educational contexts in Serbia, providing functionality for the analysis of textual content prior to its publication on internet portals and social media platforms. Key features include the automatic detection of hate speech and the generation of revised, nonhateful versions of the text while preserving its original intent.

• M.Stanojević, D.Drašković, "Razvoj infrastrukture za upravljanje Docker kontejnerima i velikim jezičkim modelima" (in Serbian; in English: "Development of Infrastructure for Managing Docker Containers and Large Language Models")

  M.Stanojević, D.Drašković, "Razvoj infrastrukture za upravljanje Docker kontejnerima i velikim jezičkim modelima", Zbornik radova 32. IKT konferencije "YU INFO 2026", Informaciono društvo Srbije, pp. 1-6, Kopaonik, Serbia, March 2026 []
  Link: View Conference Proceedings (in publishing!) | Download paper (PDF)
  Abstract: With the increasing adoption of large language models across various domains, ranging from natural language processing and intelligent assistants to the automation of complex business processes, there is a growing need for efficient, reliable, and standardized systems that enable their straightforward deployment, maintenance, and scalable management. Large language models require substantial hardware resources, specialized execution environments, and carefully planned mechanisms for task scheduling and comprehensive source code version management, which further underscores the importance of a well-designed infrastructure. At the same time, Docker containers have become one of the key components of modern software architectures. This paper places particular emphasis on system architecture, the organization and automation of processes, as well as the integration of models into complex industrial environments.