Lyubomyr Chyrun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Lyubomyr Chyrun has a total of 102 co-authors. Here are the top co-authors by number of shared publications:
- Victoria Vysotska
Lviv Polytechnic National University, Ukraine
29 shared publications Ivan Franko National University of Lviv
14 shared publicationsLviv Polytechnic National University
14 shared publications- Sofia Chyrun
Lviv Polytechnic National University
14 shared publications - Dmytro Uhryn
Yuriy Fedkovych Chernivtsi National University
9 shared publications - Yuriy A. Ushenko
Chernivtsi National University
9 shared publications
Latest Publications
- Open Access
- 24 September 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- Vol. 2 (34) , 101-128
INFORMATION SECURITY FOR SOCIAL NETWORK CYBERSPACE BY DETECTING WAYS OF DISINFORMATION SPREAD AND INAUTHENTIVE CHAT USERS' BEHAVIOR BASED ON DEEP LEARNING AND MULTIMODAL ANALYSIS
The article considers the problem of identifying sources and methods of disseminating disinformation and inauthentic behaviour among chat users in the modern information space. The relevance of the study stems from the rapid growth of digital content, the spread of social networks and chat platforms, and the intensification of information attacks, manipulative influence, and the use of automated bot networks. It is shown that traditional approaches based exclusively on the analysis of text content have significant limitations due to their inability to fully account for the behavioural characteristics of users, multimedia components, and structural relationships among participants in the information environment. The first section of the work analyses the current state of the problem of disinformation detection and investigates the main methods for automatic analysis of false content and inauthentic user behaviour. Traditional machine learning approaches, modern deep learning models, multimodal analysis methods and graph neural networks are considered. A comparative analysis of the advantages and disadvantages of existing approaches is conducted, problematic aspects are identified, and promising areas for development are outlined. In the problem statement section, a conceptual research pipeline is proposed that covers multimodal data collection and preprocessing, feature extraction, multimodal fusion, training deep learning models, building user interaction graphs, analysing information cascades, and identifying potential sources of disinformation. A mathematical model integrating text, behavioural, and graph characteristics is developed. The methods and materials section describes the mathematical apparatus of the model, which combines Transformer models for text analysis, convolutional neural networks for visual feature extraction, recurrent networks for behavioural analysis, and Graph Neural Networks for modelling the structure of user interactions. A multimodal feature fusion mechanism using attention mechanisms and combined loss functions is proposed. In the experimental part, a study was conducted using multimodal datasets comprising text messages, activity time series, and user network characteristics. The results demonstrated the effectiveness of the proposed approach in identifying potential sources of disinformation, detecting anomalous behavioural patterns, and predicting how information messages spread. Analysis of interactive distribution graphs and probability maps confirmed the system's ability to adapt to different network structures and information cascade types. The practical significance of the study lies in the possibility of using the developed approach in information space monitoring systems, cybersecurity, social networks, chat platforms and early detection systems of information threats.
- 24 September 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- Vol. 2 (34) , 101-128
Open AccessINFORMATION SECURITY FOR SOCIAL NETWORK CYBERSPACE BY DETECTING WAYS OF DISINFORMATION SPREAD AND INAUTHENTIVE CHAT USERS' BEHAVIOR BASED ON DEEP LEARNING AND MULTIMODAL ANALYSIS
The article considers the problem of identifying sources and methods of disseminating disinformation and inauthentic behaviour among chat users in the modern information space. The relevance of the study stems from the rapid growth of digital content, the spread of social networks and chat platforms, and the intensification of information attacks, manipulative influence, and the use of automated bot networks. It is shown that traditional approaches based exclusively on the analysis of text content have significant limitations due to their inability to fully account for the behavioural characteristics of users, multimedia components, and structural relationships among participants in the information environment. The first section of the work analyses the current state of the problem of disinformation detection and investigates the main methods for automatic analysis of false content and inauthentic user behaviour. Traditional machine learning approaches, modern deep learning models, multimodal analysis methods and graph neural networks are considered. A comparative analysis of the advantages and disadvantages of existing approaches is conducted, problematic aspects are identified, and promising areas for development are outlined. In the problem statement section, a conceptual research pipeline is proposed that covers multimodal data collection and preprocessing, feature extraction, multimodal fusion, training deep learning models, building user interaction graphs, analysing information cascades, and identifying potential sources of disinformation. A mathematical model integrating text, behavioural, and graph characteristics is developed. The methods and materials section describes the mathematical apparatus of the model, which combines Transformer models for text analysis, convolutional neural networks for visual feature extraction, recurrent networks for behavioural analysis, and Graph Neural Networks for modelling the structure of user interactions. A multimodal feature fusion mechanism using attention mechanisms and combined loss functions is proposed. In the experimental part, a study was conducted using multimodal datasets comprising text messages, activity time series, and user network characteristics. The results demonstrated the effectiveness of the proposed approach in identifying potential sources of disinformation, detecting anomalous behavioural patterns, and predicting how information messages spread. Analysis of interactive distribution graphs and probability maps confirmed the system's ability to adapt to different network structures and information cascade types. The practical significance of the study lies in the possibility of using the developed approach in information space monitoring systems, cybersecurity, social networks, chat platforms and early detection systems of information threats.
- Open Access
- 25 June 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- Vol. 1 (33) , 207-224
RESEARCH INTO MECHANISMS FOR DETECTING SOURCES AND WAYS OF FAKE NEWS AND PROPAGANDA DISSEMINATION IN SOCIAL NETWORKS CYBERSPACE
The article presents an information technology for detecting sources and routes of fake news and propaganda distribution in social networks and online media. The proposed approach is based on vectorization of text publications using language models, calculation of cosine similarity and clustering of messages to identify thematically similar groups. Based on the formed clusters, a chronological graph visualization is built, which allows identifying the primary sources of disinformation, key relays and the speed of its distribution. The experimental study was conducted on a combined dataset of over 2,000 posts, in which the shares of true and fake messages are relatively balanced. Interaction analysis showed that about 75% of all posts receive no more than 20 likes, while only less than 5% of posts form a “long tail” with hundreds and thousands of reactions. At the same time, disinformation posts are more likely to either remain almost unnoticed or to sharply gain abnormally high popularity in a short period of time. Analysis of distributions revealed that approximately 80% of posts have no more than 5 reposts, however, the share of fakes among the most widely shared messages is growing significantly. Platform analysis showed that reliable content prevails on web resources and authoritative media, while on social networks and messengers (in particular Telegram) the ratio of true and fake messages is close to equilibrium, and in some network’s disinformation dominates. A comparison of embedding models showed that the OpenAI model provides a clearer separation of messages in the feature space and allows detecting up to 10–11 significant clusters at an optimal cosine similarity threshold of τ ≈ 0.55–0.60, while for the local model the optimal threshold is much higher (τ ≈ 0.86–0.88). The constructed propagation graphs showed that the time interval between the initial publication of a fake and its appearance on other platforms can be from several hours to several days, and the transformation of reposts into “original” publications is a typical mechanism for hiding the source.
- 25 June 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- Vol. 1 (33) , 207-224
Open AccessRESEARCH INTO MECHANISMS FOR DETECTING SOURCES AND WAYS OF FAKE NEWS AND PROPAGANDA DISSEMINATION IN SOCIAL NETWORKS CYBERSPACE
The article presents an information technology for detecting sources and routes of fake news and propaganda distribution in social networks and online media. The proposed approach is based on vectorization of text publications using language models, calculation of cosine similarity and clustering of messages to identify thematically similar groups. Based on the formed clusters, a chronological graph visualization is built, which allows identifying the primary sources of disinformation, key relays and the speed of its distribution. The experimental study was conducted on a combined dataset of over 2,000 posts, in which the shares of true and fake messages are relatively balanced. Interaction analysis showed that about 75% of all posts receive no more than 20 likes, while only less than 5% of posts form a “long tail” with hundreds and thousands of reactions. At the same time, disinformation posts are more likely to either remain almost unnoticed or to sharply gain abnormally high popularity in a short period of time. Analysis of distributions revealed that approximately 80% of posts have no more than 5 reposts, however, the share of fakes among the most widely shared messages is growing significantly. Platform analysis showed that reliable content prevails on web resources and authoritative media, while on social networks and messengers (in particular Telegram) the ratio of true and fake messages is close to equilibrium, and in some network’s disinformation dominates. A comparison of embedding models showed that the OpenAI model provides a clearer separation of messages in the feature space and allows detecting up to 10–11 significant clusters at an optimal cosine similarity threshold of τ ≈ 0.55–0.60, while for the local model the optimal threshold is much higher (τ ≈ 0.86–0.88). The constructed propagation graphs showed that the time interval between the initial publication of a fake and its appearance on other platforms can be from several hours to several days, and the transformation of reposts into “original” publications is a typical mechanism for hiding the source.
- Open Access
- 31 May 2026
- Published by Khmelnytskyi National University in Computer Systems and Information Technologies
- No. 2,p. 214-228
SENTIMENT ANALYSIS OF PUBLIC OPINION REGARDING THE WAR IN UKRAINE BASED ON REDDIT DATA USING NLP AND MACHINE LEARNING METHODS
The article examines public opinion among Americans regarding the war in Ukraine by analysing text data from the social platform Reddit. The relevance of the work lies in the significant influence of public sentiment in the United States on the formation of foreign policy and on public support for Ukraine. The purpose of the study is to conduct a comprehensive sentiment analysis of English-language comments of Reddit users using modern methods of natural language processing (NLP) and machine learning. The work uses a dataset from the Kaggle platform that contains millions of comments on the Russian-Ukrainian war. Preprocessing of text data was carried out, including cleaning, tokenisation, lemmatisation and removal of stop words. For tone analysis, both classical approaches (VADER, TextBlob) and modern machine learning models were used, including Logistic Regression, Random Forest, SVM, XGBoost, and Naive Bayes. A hybrid approach to text vectorisation (TF-IDF with Word2Vec) was implemented. The results obtained allow us to determine the distribution of emotional assessments (positive, negative, neutral), identify thematic clusters of discussions and investigate the dynamics of changes in public sentiment over time. A comparative analysis of the effectiveness of the models based on the main quality metrics was conducted. Particular attention was paid to the specifics of Reddit discourse, including sarcasm, irony and political polarisation. The practical value of the study lies in the creation of analytical tools for monitoring public opinion, which can be used in diplomacy, politics, and the media to develop effective communication strategies.
- 31 May 2026
- Published by Khmelnytskyi National University in Computer Systems and Information Technologies
- No. 2,p. 214-228
Open AccessSENTIMENT ANALYSIS OF PUBLIC OPINION REGARDING THE WAR IN UKRAINE BASED ON REDDIT DATA USING NLP AND MACHINE LEARNING METHODS
The article examines public opinion among Americans regarding the war in Ukraine by analysing text data from the social platform Reddit. The relevance of the work lies in the significant influence of public sentiment in the United States on the formation of foreign policy and on public support for Ukraine. The purpose of the study is to conduct a comprehensive sentiment analysis of English-language comments of Reddit users using modern methods of natural language processing (NLP) and machine learning. The work uses a dataset from the Kaggle platform that contains millions of comments on the Russian-Ukrainian war. Preprocessing of text data was carried out, including cleaning, tokenisation, lemmatisation and removal of stop words. For tone analysis, both classical approaches (VADER, TextBlob) and modern machine learning models were used, including Logistic Regression, Random Forest, SVM, XGBoost, and Naive Bayes. A hybrid approach to text vectorisation (TF-IDF with Word2Vec) was implemented. The results obtained allow us to determine the distribution of emotional assessments (positive, negative, neutral), identify thematic clusters of discussions and investigate the dynamics of changes in public sentiment over time. A comparative analysis of the effectiveness of the models based on the main quality metrics was conducted. Particular attention was paid to the specifics of Reddit discourse, including sarcasm, irony and political polarisation. The practical value of the study lies in the creation of analytical tools for monitoring public opinion, which can be used in diplomacy, politics, and the media to develop effective communication strategies.
- 30 May 2026
- Published by Kharkiv National University of Radioelectronics in Bionics of Intelligence
- Vol. 2 (105) , 60-77
Information technology of multispectral video monitoring for human injuries detection and classification in emergency situations based on computer vision, convolutional neural networks and edge-computing
The article investigates approaches to automated identification of human injuries in emergency situations using multispectral video monitoring and deep learning. An intelligent system is proposed that combines visible and infrared spectral data to improve the accuracy of human detection and classification of injury types in complex environmental conditions, such as low light, smoke, or partial obstructions. The system architecture includes a multispectral video sensor module, data preprocessing and fusion modules, human detection algorithms, neural network models for feature extraction and injury classification, and a subsystem for assessing injury severity. Computer vision methods, convolutional neural networks, and edge-computing technology were used to implement data processing, which provides real-time video stream analysis without significant data transmission delays. Experimental studies were conducted on a multispectral set of images with different observation conditions and injury types. The experimental results showed that using a multispectral approach (RGB with IR) significantly increases the accuracy of human detection and damage classification compared to traditional RGB systems. The proposed model achieved an accuracy of 95,9 %, a Precision of 94,3 %, a Recall of 93,2 %, and an mAP of 0.92. The use of edge-computing technology enabled real-time video stream processing at over 25 frames per second and reduced data transmission latency to the central system. The results showed that using a multispectral approach increases the accuracy of human detection and injury classification to 96%, while ensuring stable system operation in difficult environmental conditions. The results confirm the promising application of multispectral sensors and deep learning models in creating intelligent systems to support rescue services and automate medical triage of victims.
- 30 May 2026
- Published by Kharkiv National University of Radioelectronics in Bionics of Intelligence
- Vol. 2 (105) , 60-77
Information technology of multispectral video monitoring for human injuries detection and classification in emergency situations based on computer vision, convolutional neural networks and edge-computing
The article investigates approaches to automated identification of human injuries in emergency situations using multispectral video monitoring and deep learning. An intelligent system is proposed that combines visible and infrared spectral data to improve the accuracy of human detection and classification of injury types in complex environmental conditions, such as low light, smoke, or partial obstructions. The system architecture includes a multispectral video sensor module, data preprocessing and fusion modules, human detection algorithms, neural network models for feature extraction and injury classification, and a subsystem for assessing injury severity. Computer vision methods, convolutional neural networks, and edge-computing technology were used to implement data processing, which provides real-time video stream analysis without significant data transmission delays. Experimental studies were conducted on a multispectral set of images with different observation conditions and injury types. The experimental results showed that using a multispectral approach (RGB with IR) significantly increases the accuracy of human detection and damage classification compared to traditional RGB systems. The proposed model achieved an accuracy of 95,9 %, a Precision of 94,3 %, a Recall of 93,2 %, and an mAP of 0.92. The use of edge-computing technology enabled real-time video stream processing at over 25 frames per second and reduced data transmission latency to the central system. The results showed that using a multispectral approach increases the accuracy of human detection and injury classification to 96%, while ensuring stable system operation in difficult environmental conditions. The results confirm the promising application of multispectral sensors and deep learning models in creating intelligent systems to support rescue services and automate medical triage of victims.
- 8 April 2026
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 18 (2) , 151-207
Information Technology for VR Training Evaluation with First Aid Skills Improvement to Detecting Human Resource Damage in Emergencies based on Behavioural Methods
Traditional first aid preparation methods often fail to reproduce realistic stress levels and to simulate visual difficulty in identifying lesions in critical situations. In emergencies, delays in recognising injuries or errors in protocols result in critical losses of human resources. The use of computer graphics and virtual reality technologies enables you to create a safe yet highly realistic environment for rescuers to test and improve their skills. The article presents an integrated methodological framework for assessing the effectiveness of VR first-aid training in conditions of damage to civilian infrastructure. The main focus is on developing mathematical models and algorithms to identify and evaluate the quality of rescuers' actions by analysing digital interaction signals in a virtual environment. A composite efficiency indicator is proposed that combines normalised parameters for reaction time, manipulation accuracy, stress level, and immersion. The work aims to formalise a mathematical model to assess the effectiveness of VR training in developing skills for lesion identification and first aid provision, using quantitative metrics. The study aims to identify statistically significant differences in learning speed and skill retention between groups using VR simulations and traditional methods. The project aims to validate innovative content creation methods, including mobile photogrammetry, to visualise damaged infrastructure and victim models. The study used a comprehensive approach that includes mobile photogrammetry and generative neural networks to create a library of 3D assets with varying degrees of detail. Performance score is based on composite indicator that integrates normalised data on reaction time, manipulation accuracy, error count, immersion rate. Linear mixed models, exponential approximations, and bootstrap estimation of effect stability were used to analyse hierarchical data and individual learning trajectories. The experimental part includes the use of mobile photogrammetry and generative neural networks to create realistic 3D models of affected environments and identify types of injuries (bleeding, burns, unconsciousness). To analyse the dynamics of learning and maintaining skills, models with mixed effects and exponential forgetting curves are used. The results confirm that the use of VR technologies provides a statistically significant acceleration in the development of automated skills for lesion identification and assistance compared to traditional methods. The proposed approach is a scalable tool for preparing civil and rescue services to act in critical situations. Experimental data showed that the integral performance score in the VR group increased from 0.42 0.10 to 0.76 0.08, while in the control group it increased from only 0.40 0.09 to 0.55 0.10 (p < 0.001). The largest effect was observed in the bleeding arrest scenario, where the effect size (Cohen's d) reached 2.3. The analysis of forgetting curves confirmed the superiority of VR: the skill loss rate in the VR group was 0.25, providing knowledge retention 1.8 times longer than in the control group (0.45). The study confirmed that VR simulations significantly accelerate the formation of automated behaviour patterns and reduce reaction time in extreme conditions. The proposed mathematical assessment model provides objective feedback and standardisation of the rescue training process. The results indicate the high practical value of introducing such tools into training programs for civilian and military structures to minimise losses in real emergencies.
- 8 April 2026
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 18 (2) , 151-207
Information Technology for VR Training Evaluation with First Aid Skills Improvement to Detecting Human Resource Damage in Emergencies based on Behavioural Methods
Traditional first aid preparation methods often fail to reproduce realistic stress levels and to simulate visual difficulty in identifying lesions in critical situations. In emergencies, delays in recognising injuries or errors in protocols result in critical losses of human resources. The use of computer graphics and virtual reality technologies enables you to create a safe yet highly realistic environment for rescuers to test and improve their skills. The article presents an integrated methodological framework for assessing the effectiveness of VR first-aid training in conditions of damage to civilian infrastructure. The main focus is on developing mathematical models and algorithms to identify and evaluate the quality of rescuers' actions by analysing digital interaction signals in a virtual environment. A composite efficiency indicator is proposed that combines normalised parameters for reaction time, manipulation accuracy, stress level, and immersion. The work aims to formalise a mathematical model to assess the effectiveness of VR training in developing skills for lesion identification and first aid provision, using quantitative metrics. The study aims to identify statistically significant differences in learning speed and skill retention between groups using VR simulations and traditional methods. The project aims to validate innovative content creation methods, including mobile photogrammetry, to visualise damaged infrastructure and victim models. The study used a comprehensive approach that includes mobile photogrammetry and generative neural networks to create a library of 3D assets with varying degrees of detail. Performance score is based on composite indicator that integrates normalised data on reaction time, manipulation accuracy, error count, immersion rate. Linear mixed models, exponential approximations, and bootstrap estimation of effect stability were used to analyse hierarchical data and individual learning trajectories. The experimental part includes the use of mobile photogrammetry and generative neural networks to create realistic 3D models of affected environments and identify types of injuries (bleeding, burns, unconsciousness). To analyse the dynamics of learning and maintaining skills, models with mixed effects and exponential forgetting curves are used. The results confirm that the use of VR technologies provides a statistically significant acceleration in the development of automated skills for lesion identification and assistance compared to traditional methods. The proposed approach is a scalable tool for preparing civil and rescue services to act in critical situations. Experimental data showed that the integral performance score in the VR group increased from 0.42 0.10 to 0.76 0.08, while in the control group it increased from only 0.40 0.09 to 0.55 0.10 (p < 0.001). The largest effect was observed in the bleeding arrest scenario, where the effect size (Cohen's d) reached 2.3. The analysis of forgetting curves confirmed the superiority of VR: the skill loss rate in the VR group was 0.25, providing knowledge retention 1.8 times longer than in the control group (0.45). The study confirmed that VR simulations significantly accelerate the formation of automated behaviour patterns and reduce reaction time in extreme conditions. The proposed mathematical assessment model provides objective feedback and standardisation of the rescue training process. The results indicate the high practical value of introducing such tools into training programs for civilian and military structures to minimise losses in real emergencies.
- 1 April 2026
- Published by Lviv Polytechnic National University in Information and communication technologies, electronic engineering
- Vol. 6 (1) , 64-74
INFORMATION TECHNOLOGY FOR IDENTIFYING THE UKRAINIAN-LANGUAGE TEXT CONTENT TOXICITY ON YOUTUBE AND GOOGLE PLAY BASED ON MACHINE LEARNING METHODS
The article presents the results of the development and experimental study of specialised information technology for automatically identifying the toxicity of Ukrainian-language text posted on YouTube and Google Play. The relevance of the work stems from the rapid growth of user comments in the digital environment, which is accompanied by the spread of aggressive, discriminatory, and hostile statements that negatively affect the quality of online communication and users' psychological comfort. Particular attention is paid to the shortage of marked corpora in Ukrainian and the limited applicability of pre-trained transformer models without additional adaptation to the morphological and semantic specifics of the Ukrainian segment. The purpose of the study is to design and implement a universal machine learning pipeline for binary classification of text messages based on toxicity, accounting for the peculiarities of Ukrainian-language digital discourse. To form a training sample, an open dataset from the Kaggle platform and the author's collections of comments from YouTube and Google Play were combined, ensuring representativeness and diversity of stylistic constructions. The general corpus contains both overtly aggressive and contextually hidden manifestations of toxic behaviour. Within the proposed information technology framework, a complete data processing cycle has been implemented: preprocessing of text, normalisation, extraction of stop words, vectorisation using TF-IDF, class balancing using the SMOTE method, and hyperparameter optimisation using GridSearchCV. A comparative analysis of six machine learning models was conducted: Logistic Regression, SVM with RBF kernel, Linear SVC, SGD Classifier, Random Forest, and Gradient Boosting. Performance was evaluated using the Accuracy, Precision, Recall, F1-score, and AUC-ROC metrics. Experimental results showed that linear models perform competitively in classifying high-dimensional, sparse text vectors. In particular, Linear SVC and SGD Classifier provided the best balance between accuracy and completeness, while ensemble methods showed higher accuracy by reducing sensitivity to toxic examples. The results confirm the feasibility of using classical machine learning algorithms under limited-resource conditions for the Ukrainian language and serve as a baseline for further integration of deep and multimodal approaches. The proposed information technology can be used as a component of automated real-time content moderation systems, thereby increasing information security and creating a safer online environment for Ukrainian-speaking users.
- 1 April 2026
- Published by Lviv Polytechnic National University in Information and communication technologies, electronic engineering
- Vol. 6 (1) , 64-74
INFORMATION TECHNOLOGY FOR IDENTIFYING THE UKRAINIAN-LANGUAGE TEXT CONTENT TOXICITY ON YOUTUBE AND GOOGLE PLAY BASED ON MACHINE LEARNING METHODS
The article presents the results of the development and experimental study of specialised information technology for automatically identifying the toxicity of Ukrainian-language text posted on YouTube and Google Play. The relevance of the work stems from the rapid growth of user comments in the digital environment, which is accompanied by the spread of aggressive, discriminatory, and hostile statements that negatively affect the quality of online communication and users' psychological comfort. Particular attention is paid to the shortage of marked corpora in Ukrainian and the limited applicability of pre-trained transformer models without additional adaptation to the morphological and semantic specifics of the Ukrainian segment. The purpose of the study is to design and implement a universal machine learning pipeline for binary classification of text messages based on toxicity, accounting for the peculiarities of Ukrainian-language digital discourse. To form a training sample, an open dataset from the Kaggle platform and the author's collections of comments from YouTube and Google Play were combined, ensuring representativeness and diversity of stylistic constructions. The general corpus contains both overtly aggressive and contextually hidden manifestations of toxic behaviour. Within the proposed information technology framework, a complete data processing cycle has been implemented: preprocessing of text, normalisation, extraction of stop words, vectorisation using TF-IDF, class balancing using the SMOTE method, and hyperparameter optimisation using GridSearchCV. A comparative analysis of six machine learning models was conducted: Logistic Regression, SVM with RBF kernel, Linear SVC, SGD Classifier, Random Forest, and Gradient Boosting. Performance was evaluated using the Accuracy, Precision, Recall, F1-score, and AUC-ROC metrics. Experimental results showed that linear models perform competitively in classifying high-dimensional, sparse text vectors. In particular, Linear SVC and SGD Classifier provided the best balance between accuracy and completeness, while ensemble methods showed higher accuracy by reducing sensitivity to toxic examples. The results confirm the feasibility of using classical machine learning algorithms under limited-resource conditions for the Ukrainian language and serve as a baseline for further integration of deep and multimodal approaches. The proposed information technology can be used as a component of automated real-time content moderation systems, thereby increasing information security and creating a safer online environment for Ukrainian-speaking users.
- Open Access
- 26 March 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- p. 654
INFORMATION TECHNOLOGY FOR UKRAINIAN-LANGUAGE FAKE NEWS DETECTION IN SOCIAL NETWORKS CYBERSPACE BASED ON MACHINE LEARNING AND NATURAL LANGUAGE PROCESSING METHODS
The article addresses the problem of automated fake news detection in the Ukrainian-language information space, which has become particularly relevant under conditions of hybrid warfare and the intensive use of disinformation as a tool of information influence. The aim of the study is to develop and experimentally evaluate an effective system for classifying Ukrainian-language news texts using natural language processing and machine learning methods. The introductory section substantiates the relevance of the research topic, defines the object and subject of the study, and formulates its goal and main objectives. The related work section provides an overview of existing approaches to fake news detection, including classical machine learning algorithms, deep neural networks, and transformer-based models, and highlights their limitations in the context of the Ukrainian language. The theoretical section systematizes methods of automated text analysis and identifies linguistic features of Ukrainian news content, as well as challenges related to the lack of large annotated corpora and language-specific resources. The methodology section describes the complete research pipeline, including data collection, text preprocessing, cleaning, tokenization, and lemmatization using tools adapted for Ukrainian, as well as the construction of semantic vector representations based on the Sentence-BERT model. Several machine learning classifiers, namely Logistic Regression, Random Forest, Support Vector Machine, and XGBoost, were implemented and compared using cross-validation and hyperparameter optimization. The results section presents an experimental evaluation of the models using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The findings demonstrate that the Support Vector Machine model achieves the best performance, reaching a classification accuracy of 93.2% even under class imbalance conditions. Computational efficiency and runtime characteristics of the proposed approach are also analysed, along with its potential for practical deployment. The conclusions summarize the main outcomes of the study, confirm that the research objectives have been achieved, and outline directions for future work, including adaptation to a multilingual environment and improving model interpretability. The proposed system can be applied in media monitoring, fact-checking initiatives, and tools aimed at strengthening information security.
- 26 March 2026
- Published by Borys Grinchenko Kyiv University in Cybersecurity: Education, Science, Technique
- p. 654
Open AccessINFORMATION TECHNOLOGY FOR UKRAINIAN-LANGUAGE FAKE NEWS DETECTION IN SOCIAL NETWORKS CYBERSPACE BASED ON MACHINE LEARNING AND NATURAL LANGUAGE PROCESSING METHODS
The article addresses the problem of automated fake news detection in the Ukrainian-language information space, which has become particularly relevant under conditions of hybrid warfare and the intensive use of disinformation as a tool of information influence. The aim of the study is to develop and experimentally evaluate an effective system for classifying Ukrainian-language news texts using natural language processing and machine learning methods. The introductory section substantiates the relevance of the research topic, defines the object and subject of the study, and formulates its goal and main objectives. The related work section provides an overview of existing approaches to fake news detection, including classical machine learning algorithms, deep neural networks, and transformer-based models, and highlights their limitations in the context of the Ukrainian language. The theoretical section systematizes methods of automated text analysis and identifies linguistic features of Ukrainian news content, as well as challenges related to the lack of large annotated corpora and language-specific resources. The methodology section describes the complete research pipeline, including data collection, text preprocessing, cleaning, tokenization, and lemmatization using tools adapted for Ukrainian, as well as the construction of semantic vector representations based on the Sentence-BERT model. Several machine learning classifiers, namely Logistic Regression, Random Forest, Support Vector Machine, and XGBoost, were implemented and compared using cross-validation and hyperparameter optimization. The results section presents an experimental evaluation of the models using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The findings demonstrate that the Support Vector Machine model achieves the best performance, reaching a classification accuracy of 93.2% even under class imbalance conditions. Computational efficiency and runtime characteristics of the proposed approach are also analysed, along with its potential for practical deployment. The conclusions summarize the main outcomes of the study, confirm that the research objectives have been achieved, and outline directions for future work, including adaptation to a multilingual environment and improving model interpretability. The proposed system can be applied in media monitoring, fact-checking initiatives, and tools aimed at strengthening information security.
- Open Access
- 26 March 2026
- Published by Khmelnytskyi National University in Computer Systems and Information Technologies
- No. 1,p. 88-103
CROSS-LINGUAL TRANSFORMER-BASED SCREENING OF POST-TRAUMATIC STRESS DISORDER BASED ON COMPARATIVE ANALYSIS OF BERT AND XLM-ROBERTA WITH MACHINE TRANSLATION ADAPTATION FOR UKRAINIAN LANGUAGE
The paper presents a comprehensive study on automated screening of post-traumatic stress disorder (PTSD) using transformer-based natural language processing models in a cross-lingual setting. The research aims to evaluate the feasibility of deploying an intelligent PTSD detection system in Ukraine under conditions of limited, localised training data. A balanced corpus of 4,822 text records was constructed by aggregating publicly available PTSD-related datasets, including 2,042 PTSD-positive texts and 2,780 control texts representing neutral content and other psychological conditions. The study compares the performance of the English-language BERT (bert-base-uncased) model and the multilingual XLM-RoBERTa (xlm-roberta-base) model applied to a Ukrainian-language corpus generated via machine translation using the Google Translate API and large language models for complex structures. Word cloud visualisation and semantic analysis confirmed preservation of core psychological markers during translation. Experimental results demonstrate high predictive performance for both architectures. The English-language model achieved an Accuracy of 0.90 and an ROC-AUC of 0.962. In contrast, the Ukrainian-language model achieved an Accuracy of 0.85 and an ROC-AUC of 0.940, significantly outperforming existing Ukrainian multi-class stress detection models (Accuracy ~0.45) and exceeding standard multilingual mental health benchmarks (0.78–0.82), establishing a robust state-of-the-art baseline for Ukrainian clinical NLP. Importantly, Recall remained identical (0.88) across both language settings, indicating strong sensitivity to PTSD markers despite translation-induced lexical noise. The minimal AUC degradation (2.3%) confirms the robustness of transformer architectures to cross-lingual adaptation. The findings validate the viability of combining machine translation with multilingual transformers for the rapid deployment of mental health screening systems in low-resource language environments. The proposed pipeline enables scalable and cost-effective digital PTSD monitoring while maintaining clinically relevant diagnostic sensitivity.
- 26 March 2026
- Published by Khmelnytskyi National University in Computer Systems and Information Technologies
- No. 1,p. 88-103
Open AccessCROSS-LINGUAL TRANSFORMER-BASED SCREENING OF POST-TRAUMATIC STRESS DISORDER BASED ON COMPARATIVE ANALYSIS OF BERT AND XLM-ROBERTA WITH MACHINE TRANSLATION ADAPTATION FOR UKRAINIAN LANGUAGE
The paper presents a comprehensive study on automated screening of post-traumatic stress disorder (PTSD) using transformer-based natural language processing models in a cross-lingual setting. The research aims to evaluate the feasibility of deploying an intelligent PTSD detection system in Ukraine under conditions of limited, localised training data. A balanced corpus of 4,822 text records was constructed by aggregating publicly available PTSD-related datasets, including 2,042 PTSD-positive texts and 2,780 control texts representing neutral content and other psychological conditions. The study compares the performance of the English-language BERT (bert-base-uncased) model and the multilingual XLM-RoBERTa (xlm-roberta-base) model applied to a Ukrainian-language corpus generated via machine translation using the Google Translate API and large language models for complex structures. Word cloud visualisation and semantic analysis confirmed preservation of core psychological markers during translation. Experimental results demonstrate high predictive performance for both architectures. The English-language model achieved an Accuracy of 0.90 and an ROC-AUC of 0.962. In contrast, the Ukrainian-language model achieved an Accuracy of 0.85 and an ROC-AUC of 0.940, significantly outperforming existing Ukrainian multi-class stress detection models (Accuracy ~0.45) and exceeding standard multilingual mental health benchmarks (0.78–0.82), establishing a robust state-of-the-art baseline for Ukrainian clinical NLP. Importantly, Recall remained identical (0.88) across both language settings, indicating strong sensitivity to PTSD markers despite translation-induced lexical noise. The minimal AUC degradation (2.3%) confirms the robustness of transformer architectures to cross-lingual adaptation. The findings validate the viability of combining machine translation with multilingual transformers for the rapid deployment of mental health screening systems in low-resource language environments. The proposed pipeline enables scalable and cost-effective digital PTSD monitoring while maintaining clinically relevant diagnostic sensitivity.
- Open Access
- 1 January 2026
- Published by Lviv Polytechnic National University in Ukrainian Journal of Information Technology
- Vol. 8 (1) , 16-31
INFORMATION TECHNOLOGY FOR IDENTIFYING PROPAGANDA IN TIKTOK COMMENTS BASED ON NLP AND DEEP LEARNING
The article investigates the current scientific problem of automated detection of propaganda influence in short text comments of users of the TikTok social network, which operates in conditions of hybrid warfare and intensive disinformation campaigns. A hybrid model for detecting propaganda content has been developed, which integrates deep contextual representations of the text (transformer-based contextual representations) based on the BERT architecture with an additional vector of semiotic and structural features (number of emojis, repetition of symbols, use of caps lock). The final classification block of the hybrid model is implemented as a two-layer fully connected perceptron (MLP), which takes as input the concatenation of the [CLS]-token vector of the transformer model (768 dimensions) and the vector of additional semiotic features (10–20 dimensions). A methodological approach is proposed that combines classification with a teacher and elements of cluster analysis (k-means clustering) for effective work in conditions of limited marked-up data corpora (low-resource languages) for the Ukrainian and Russian languages. It has been found that the integration of additional lexical and graphic markers allows the model to better identify hidden ideological narratives, which often do not contain direct aggression. It was found that the implementation of the hybrid architecture provides a statistically significant improvement in quality metrics: the F1-score increased by 3–5% compared to the basic transformer models, and the area under the curve (ROC-AUC) value increased by 0.04–0.06. Specific differences in the functioning of the model for different language segments were identified: for the Ukrainian-speaking corpus, higher accuracy (Precision) was recorded, while for the Russian-speaking one, higher completeness of detection (Recall) of propaganda messages. The effectiveness of the developed system was studied on real control examples (use cases), which confirmed the model’s ability to detect up to 80% of propaganda content in thematic videos about Crimea and other socio-political issues. The information space monitoring tools were improved through the implementation of an application interface in the form of a Telegram bot, which provides operational analysis of TikTok comments in real time. Prospects for further research lie in expanding multimodal analysis for simultaneous processing of text, video, and audio tracks of a social network.
- 1 January 2026
- Published by Lviv Polytechnic National University in Ukrainian Journal of Information Technology
- Vol. 8 (1) , 16-31
Open AccessINFORMATION TECHNOLOGY FOR IDENTIFYING PROPAGANDA IN TIKTOK COMMENTS BASED ON NLP AND DEEP LEARNING
The article investigates the current scientific problem of automated detection of propaganda influence in short text comments of users of the TikTok social network, which operates in conditions of hybrid warfare and intensive disinformation campaigns. A hybrid model for detecting propaganda content has been developed, which integrates deep contextual representations of the text (transformer-based contextual representations) based on the BERT architecture with an additional vector of semiotic and structural features (number of emojis, repetition of symbols, use of caps lock). The final classification block of the hybrid model is implemented as a two-layer fully connected perceptron (MLP), which takes as input the concatenation of the [CLS]-token vector of the transformer model (768 dimensions) and the vector of additional semiotic features (10–20 dimensions). A methodological approach is proposed that combines classification with a teacher and elements of cluster analysis (k-means clustering) for effective work in conditions of limited marked-up data corpora (low-resource languages) for the Ukrainian and Russian languages. It has been found that the integration of additional lexical and graphic markers allows the model to better identify hidden ideological narratives, which often do not contain direct aggression. It was found that the implementation of the hybrid architecture provides a statistically significant improvement in quality metrics: the F1-score increased by 3–5% compared to the basic transformer models, and the area under the curve (ROC-AUC) value increased by 0.04–0.06. Specific differences in the functioning of the model for different language segments were identified: for the Ukrainian-speaking corpus, higher accuracy (Precision) was recorded, while for the Russian-speaking one, higher completeness of detection (Recall) of propaganda messages. The effectiveness of the developed system was studied on real control examples (use cases), which confirmed the model’s ability to detect up to 80% of propaganda content in thematic videos about Crimea and other socio-political issues. The information space monitoring tools were improved through the implementation of an application interface in the form of a Telegram bot, which provides operational analysis of TikTok comments in real time. Prospects for further research lie in expanding multimodal analysis for simultaneous processing of text, video, and audio tracks of a social network.
- Open Access
- 8 October 2025
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 17 (5) , 117-164
Sentiment Analysing and Visualising PublicOpinion on Political Figures across YouTube andTwitter Using NLP and Machine Learning
The study is devoted to the analysis of public sentiment towards Ukrainian political figures based on comments on social media, in particular, YouTube and Twitter. The work aims to identify differences in the perception of political leaders and to understand how the platform affects the tone of statements. The main research question is to determine how public opinion about politicians in Ukraine differs between YouTube and Twitter during the full-scale war. To do this, a corpus of comments and tweets from 2022 to 2023 was collected, which went through pre-processing stages (including cleaning up slang and spelling mistakes). The article presents the results of a comprehensive analysis of public opinion on five public figures of Ukraine (S. Prytula, P. Poroshenko, V. Zelensky, S. Sternenko, A. Yermak) based on data from the social networks YouTube and Twitter. For data collection, the YouTube Data API and the Apify platform were used, a corpus of Ukrainian-language comments and tweets was collected and processed, which went through the stages of purification, normalisation and lemmatisation, taking into account slang, surzhyk and spelling mistakes. The sentiment analysis model, built on the basis of multilingual-e5-base embeddings and the XGBClassifier algorithm, showed an accuracy of 89.4%, macro-F1 of 88.7%, and a weighted F1 of 89.1%. Sentiment distribution analysis revealed that, on average, 42% of messages were positive, 36% were negative, and 22% were neutral. Twitter had a higher share of negative statements (up to 40%), while YouTube had a predominance of positive sentiment (up to 47%). The results indicate differences in the perception of public figures on different platforms and confirm the
- 8 October 2025
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 17 (5) , 117-164
Open AccessSentiment Analysing and Visualising PublicOpinion on Political Figures across YouTube andTwitter Using NLP and Machine Learning
The study is devoted to the analysis of public sentiment towards Ukrainian political figures based on comments on social media, in particular, YouTube and Twitter. The work aims to identify differences in the perception of political leaders and to understand how the platform affects the tone of statements. The main research question is to determine how public opinion about politicians in Ukraine differs between YouTube and Twitter during the full-scale war. To do this, a corpus of comments and tweets from 2022 to 2023 was collected, which went through pre-processing stages (including cleaning up slang and spelling mistakes). The article presents the results of a comprehensive analysis of public opinion on five public figures of Ukraine (S. Prytula, P. Poroshenko, V. Zelensky, S. Sternenko, A. Yermak) based on data from the social networks YouTube and Twitter. For data collection, the YouTube Data API and the Apify platform were used, a corpus of Ukrainian-language comments and tweets was collected and processed, which went through the stages of purification, normalisation and lemmatisation, taking into account slang, surzhyk and spelling mistakes. The sentiment analysis model, built on the basis of multilingual-e5-base embeddings and the XGBClassifier algorithm, showed an accuracy of 89.4%, macro-F1 of 88.7%, and a weighted F1 of 89.1%. Sentiment distribution analysis revealed that, on average, 42% of messages were positive, 36% were negative, and 22% were neutral. Twitter had a higher share of negative statements (up to 40%), while YouTube had a predominance of positive sentiment (up to 47%). The results indicate differences in the perception of public figures on different platforms and confirm the