Fulltext search in archive

Background: Environmental monitoring and data collection critically depend on wireless sensor networks (WSN). However, the limited battery life of these sensor nodes significantly affects the operational lifetime of the network. This often results in challenges related to energy efficiency and clustering tasks due to dynamic network topologies and limited resources.
Objective: This paper aims to improve clustering efficiency and extend the network lifetime of WSN by proposing a hybrid model.
Methods: The proposed framework was evaluated in terms of energy consumption, network lifetime, Packet Delivery Ratio (PDR), and clustering efficiency through extensive simulations. Modified Salp Swarm Algorithm (MSSA)-Deep Stacked Sparse Autoencoder (DSSA)-K-means is a promising solution for next-generation WSN in smart environments and industrial Internet of Things (IoT) applications because experimental results show that it achieves higher clustering accuracy, energy economy, and robustness against network failures than current methods.
Results: Extensive simulations were conducted to evaluate the proposed framework based on energy consumption, network lifetime, PDR, and clustering efficiency. The experimental results demonstrate that the proposed MSSA-DSSA-K-means model achieves higher clustering accuracy, energy economy, and robustness against network failures compared to existing methods.
Conclusion: The paper concludes that this hybrid model is a promising solution for next-generation WSN in smart environments and industrial IoT applications.

Background: Ubiquitous learning environments aim to provide personalized and context-aware educational resources; however, traditional recommendation systems often fall short in meeting these dynamic learner needs.Objective: This study develops and evaluates a chatbot-based recommendation system that uses generative AI and prompt engineering techniques to enhance recommendation accuracy and user engagement in ubiquitous learning contexts.Methods: A ChatGPT-powered chatbot was implemented using few-shot prompting and dynamic context integration to deliver personalized, real-time educational support. The system was deployed using an intuitive Gradio interface, facilitating user accessibility and seamless interaction across varied learning scenarios. A tailored evaluation dataset was constructed to capture diverse user interactions and the system was tested through real-world case studies and user feedback metrics, including task success rates, response times and satisfaction ratings.Results: The chatbot achieved an 85% overall task success rate, a 70% success rate in context-aware tasks and an 80% user satisfaction rating, with most users assigning scores of 4 or 5 on a 5-point scale.Conclusion: The findings demonstrate that the proposed solution outperforms traditional systems in delivering personalized, adaptive and context-aware educational recommendations, underscoring the transformative potential of generative AI in advancing learner-centred ubiquitous learning environments.

Background: Recommendation systems are essential for personalized user experiences but struggle to balance accuracy and efficiency.Objective: This paper presents ResNetMF, an innovative hybrid framework designed to address these limitations by combining the strengths of matrix factorization (MF) and deep residual networks (ResNet). Matrix factorization excels at capturing explicit linear relationships between users and items, while ResNet is employed to model non-linear residuals.Methods: By focusing on refining the baseline MF output through incremental improvements, ResNetMF minimizes redundant computations and significantly enhances recommendation accuracy. The unique architecture of the framework allows it to capture and represent both linear and non-linear relationships between users and items, ensuring robust and scalable performance. Extensive experiments conducted on the widely used MovieLens dataset demonstrate the superiority of ResNetMF over existing methods.Results: Specifically, it achieves a minimum improvement of 7.95% in root mean square error compared to neural collaborative filtering and outperforms other state-of-the-art techniques in key metrics such as precision, recall and training efficiency. These results highlight the ability of ResNetMF to deliver highly accurate recommendations while maintaining computational efficiency, making it an efficient approach to real-world application of recommendation systems.Conclusion: By addressing the dual challenges of accuracy and efficiency, ResNetMF offers a balanced and scalable approach to personalized recommendation systems.

Background: Virtual reality (VR) integrated with internet of things (IoT) wearable devices offers innovative approaches to mental health interventions by enabling real-time physiological monitoring during immersive therapeutic experiences.Objective: This study aims to evaluate the effectiveness of VR therapeutic games in identifying and measuring emotional responses through physiological signals (heart rate and galvanic skin response) and to classify these responses using machine learning.Methods: We conduct experiments with 103 participants (aged 6–57 years) using FearTherapy, a custom VR game featuring interactions with four animals (Hermit, Bee, Wolf, Spider). Physiological data are collected using Samsung Galaxy Watch 5 for heart rate and Arduino Uno with a galvanic skin response (GSR) sensor. After preprocessing, 55 valid sessions remain for analysis. Individual baseline heart rate values are established and random forest classification with grid search optimization and 10-fold cross-validation is performed.Results: GSR emerges as the most influential feature for classifying emotional states, followed by heart rate difference and baseline reference values. Highest emotional arousal occurs during Spider and Bee interactions. The random forest model achieves 69% cross-validation accuracy and 81% test set accuracy. The model performs well overall but encounters challenges distinguishing Bee from Hermit and Wolf from Spider, suggesting overlapping emotional states.Conclusion: VR therapeutic games combined with IoT physiological monitoring can effectively measure and classify emotional responses. The findings support development of personalized, emotionally adaptive therapeutic interventions for anxiety and phobia treatment, emphasizing the importance of individualized baseline measurements for accurate emotional state assessment.

Background: Zimbabwe has implemented the Impilo electronic health record (EHR) system since 2016 to manage the health system electronically, gather strategic information and reduce manual documentation burden.Objective: We evaluated the capacity of decentralized structures to effectively use the Impilo EHR platform, identify training needs and challenges and provide recommendations for enhancing its effectiveness and support for integrated people-centred services at the primary healthcare level.Methods: We conducted a cross-sectional, mixed-method design, applying the COM-B (Capability, Opportunity, Motivation and Behaviour) model of behavioural change. Forty-five purposively selected healthcare workers (nurses, data entry clerks, receptionists, pharmacy staff, laboratory technicians and primary counsellors) from ten healthcare facilities in Harare and Bulawayo were included in this study. Interviews were transcribed, translated and manually coded for thematic analysis using the COM-B constructs.Results: Health workers had satisfactory skills for using the Impilo EHR system but lacked troubleshooting abilities. The capacity building did not equip users with the necessary programme-specific skills. Problems such as internet connectivity, power backup, human resource shortages, interoperability issues and lack of editing rights hindered usage. The EHR system integrated primary health services but struggled with interoperability with other software and lacked data aggregation servers, limiting its effectiveness. Leadership support and user involvement were missed opportunities to enhance performance.Conclusion: This study provided key insights into the implementation of the Impilo EHR system in Zimbabwe. The system empowers healthcare professionals with timely information, improving decision-making and patient care. However, problems such as module issues, knowledge gaps, internet connectivity, interoperability, human resource shortages and power constraints hinder its full potential. We recommend addressing these handicaps, enhancing leadership support, integrating EHR usage into performance appraisals and improving system integration with other platforms to enhance accuracy and reliability.

Background: Maintaining high motivation in robot-led educational activities is challenging when interactions rely solely on verbal communication. Incorporating multimodal feedback combining gestures, sounds and music may provide a richer and more engaging learning experience.Objective: This study aims to examine whether integrating multimodal feedback with a real-time, fair first-responder detection system in a robot-led quiz game enhances students’ intrinsic motivation and engagement compared to a verbal-only, sequential turn-taking interaction.Methods: A two-group experiment was conducted with 48 university students randomly assigned to two groups. The experimental group interacted with a Pepper robot using buzzer-based competition and synchronized multimodal feedback (gestures, sounds and music), while the control group experienced verbal-only interaction with sequential turn-taking. After the session, participants completed the Intrinsic Motivation Inventory questionnaire covering five subscales: interest/enjoyment, perceived competence, effort/importance, perceived choice and pressure/tension. Statistical analyses included t-tests, 95% confidence intervals and effect sizes (Cohen’s d) to compare group differences.Results: The experimental group reported significantly higher scores in interest/enjoyment (p < 0.001, d = 3.11), perceived competence (p < 0.001, d = 2.28), effort/importance (p < 0.001, d = 4.59) and perceived choice (p = 0.048, d = 0.93). Pressure/tension scores were also higher (p < 0.001, d = 1.95), reflecting the excitement and mild stress of competitive gameplay.Conclusion: Multimodal feedback combined with fair first-responder detection substantially enhances intrinsic motivation and engagement in robot-led learning environments. While competitive pressure increases tension, it also appears to stimulate focus and effort. These findings highlight the potential of multimodal, fair and interactive robot systems for creating more dynamic and emotionally engaging educational experiences.

The article focuses on examples of the presence of the banking and telecommunications sector on Google+. In the first part investigated set is described, and then the individual profiles are discussed from the perspective of the ways of their use and also behaving of company representatives is mentioned. In the second part of the article chosen posts of important individuals on Google+ related to the investigated sectors are discussed. The difference was found in the ways how companies use their profiles, in the posting frequency, communication styles and goals, as well as in the form of responses to critical posts and comments. Also, the examples of the contributions of significant individuals were described where the response of the representative contributed to the further conflict escalation and spread through social network.

Background: Large Language Models (LLMs) have transformed research and industry applications; however, cloud deployment decisions remain complex and poorly documented, particularly for academic researchers operating under budget constraints. Systematic guidance on infrastructure selection for LLM-based research is limited.Objective: This study provides a comprehensive empirical evaluation of cloud-based LLM deployment architectures, examining inference efficiency, serverless platform availability, and architectural trade-offs across major cloud providers to deliver actionable guidance for budget-constrained researchers.Methods: The author evaluated 32 open-source LLMs ranging from 0.6 billion to 1 trillion parameters across serverless and Bring Your Own Container (BYOC) deployment configurations. Using the Belebele benchmark, we analyzed cost–efficiency relationships, serverless platform availability, and metrics exposure across Amazon SageMaker, Amazon Bedrock, Azure Serverless, and Hugging Face–compatible providers.Results: Model performance follows a logarithmic scaling relationship with parameter count (R²=0.727) and deployment cost (R²=0.639). Models in the 30–50B parameter range achieve 85–90% of maximum accuracy at a fraction of the cost of frontier models. However, serverless availability remains fragmented: only 34.4% of examined models are accessible via serverless endpoints, with minimal cross-platform redundancy (6.2%). Deployment architecture introduces a fundamental trade-off: serverless platforms expose 71% fewer metrics than BYOC approaches while eliminating infrastructure management overhead and idle costs.Conclusion: These findings provide practical guidance for researchers selecting cloud infrastructure under budget constraints. Models in the 7–14B range offer optimal cost efficiency, while the 30–50B range maximizes accuracy per dollar for demanding tasks. The results also challenge the prevailing emphasis on ever-larger models, as diminishing returns become substantial beyond 30B parameters. Persistent gaps in serverless availability and observability highlight the need for greater standardization in cloud platforms.

Background: Land records have traditionally derived their credibility from a central database of local government records, with copies issued to land owners. Physical records are the only credible source of any information related to land ownership that has been in existence for a long time. However, physical records are prone to manipulation and fraud. Recently, some academic research has begun to address the potential use of blockchain technology to improve the security and reliability of land registration processes. Objective: The purpose of the present work is to propose an architecture for blockchain-based access control for distribution, ensuring information privacy. We take advantage of the benefits of blockchain technology in improving land record management while granting access to electronic data through user permissions. Methods: This approach replicates cryptographic primitives, while smart contracts are used to assist land record owners and users in interacting with each other using the Ethereum blockchain in the proposed system. The approach includes performance evaluation by the execution of a smart contract and security analysis to check the system robustness. Results: The performance evaluation and security analysis prove the proposed blockchain architecture to be secure and feasible for practical implementation in managing land records. Conclusion: The research proves how the application of blockchain technology can significantly enhance both security and reliability in land registration processes, giving credibility to tamper-resistant systems for maintaining information about land ownership.

Background: Researchers are using artificial intelligence (AI) tools in academic writing. However, their use may compromise the integrity and originality of the work. Hence, AI text detection tools have come to increase transparency. Objective: This study aims to evaluate the accuracy of AI text detection tools in recognizing human-written and AI-written abstracts in library and information science (LIS).Methods: Seven Persian academic journals in LIS were selected. ZeroGPT and GPTZero as AI text detectors were used. AI-generated abstracts were produced by AI chatbots (ChatGPT 4.0, DeepSeek and Qwen).Results: Despite performing strongly in detecting AI-generated text, especially from models such as DeepSeek and Qwen, ZeroGPT and GPTZero struggle to accurately identify human-written content, resulting in high false positive rates and raising concerns about their reliability.Conclusion: The findings highlight the need for culturally and linguistically inclusive AI detection tools, as current systems such as ZeroGPT and GPTZero show limitations in diverse language contexts, underscoring the importance of improved algorithms and human-involved evaluation to ensure fairness and reliability in academic settings.

Background: The rapid development of high-speed railway (HSR) systems requires advanced and reliable communication infrastructure to support operational safety, passenger services and intelligent transportation functions. Despite growing attention, comprehensive reviews of scientific developments in HSR communication systems remain limited.Objective: This article seeks to explore the development, thematic landscape and prospective directions of HSR communication studies through bibliometric analysis, emphasizing the identification of core technologies, prevailing research trends and promising avenues for future investigation.Methods: A total of 352 articles published between 2005 and 2024 were retrieved from the Scopus database to examine research development in the field. The dataset was cleaned using OpenRefine and analysed through keyword co-occurrence techniques in VOSviewer and Biblioshiny. This analysis identified thematic clusters, temporal trends and core technologies shaping the research domain.Results: The study highlights four major research clusters: artificial intelligence (AI) and adaptive communication technologies, fifth-generation (5G) network architecture and mobility solutions, signal processing and quality of service (QoS) optimization and channel modelling with propagation characteristics. Massive multiple-input multiple-output (massive MIMO) and millimetre-wave (mmWave) technologies emerge as key enablers for addressing high-mobility challenges. Furthermore, the findings reveal a growing integration of AI, edge computing and real-time communication protocols in recent research.Conclusion: This overview offers a macro-level perspective on the scientific landscape of HSR communication studies. The findings underscore the growing adoption of adaptive, intelligent and energy-efficient technologies, providing strategic guidance for future scholarly work and policymaking in advancing next-generation railway communication systems.

Background: Traditional expert systems in education rely on static knowledge bases and rule-based logic, limiting their ability to adapt to the diverse and evolving needs of students. Recent advancements in artificial intelligence and psychological profiling offer new pathways for building personalized support systems.Objective: This study presents a data science framework for developing adaptive expert systems that personalize support delivery using dynamic psychological profiling, with a focus on the higher education domain.Methods: The proposed system integrates five core components: MIND (a multimodal information orchestrator), UEX (expert system knowledge base), ULM (domain-specific large language model), PAGE (a personality-adaptive generative engine), and SYNAPSE (a dynamic profiling module). Psychological personalization is achieved using a multimodel Myers-Briggs Type Indicator (MBTI) classifier developed and validated in a separate study. In this paper, the classifier is operationalized within the system to enable real-time profiling. System behavior is driven by multimodal data fusion and continuously updated based on user interactions. The evaluation focuses on the impact of MBTI-based personalization on user satisfaction, assessed through a controlled survey comparing generic and personalized system responses.Results: In an experiment involving 70 participants with identified MBTI profiles, 79% preferred responses generated by PAGE (psychologically personalized) over generic outputs. This preference was consistent across most MBTI types, indicating the broad applicability of personalization. Minor deviations were observed for types with a preference for concise communication, suggesting variability in personalization effectiveness.Conclusion: The findings demonstrate that embedding psychological profiling into expert system workflows enhances perceived relevance and engagement of system responses. This adaptive framework enables real-time personalization through data-driven profiling, and its modular design allows for deployment across multiple domains. The system’s architecture establishes a scalable and behaviorally grounded foundation for next-generation educational support systems.

Background: Artificial intelligence (AI) is rapidly transforming human resource management (HRM) by automating essential functions such as recruitment, employee performance evaluation and workforce planning. Despite the growing adoption of AI-driven tools, organizations face numerous challenges, including resistance from HR professionals, ethical concerns and data privacy issues. This transformation has sparked significant academic interest, yet several gaps remain in understanding how AI affects HRM practices and organizational outcomes.Objective: This study aims to explore the integration of AI in HRM by analysing its potential opportunities and challenges. Additionally, it investigates how human-AI collaboration can enhance HR functions and drive organizational performance.Methods: A systematic literature review approach was adopted, focusing on 141 recent peer-reviewed articles from Scopus-indexed journals. This review focused on studies published between 2019 and 2025. The analysis was structured around three key dimensions: AI-driven opportunities, challenges associated with AI adoption and its transformative impact on HRM practices.Results: The findings reveal that AI offers significant benefits in HRM, such as improving efficiency, reducing bias and enhancing employee engagement. However, challenges remain regarding ethical decision-making, data security and maintaining human interaction in HR processes. The study also highlights the importance of augmenting, rather than replacing, human roles with AI tools to achieve optimal outcomes.Conclusion: AI has the potential to reshape HRM by streamlining processes and enhancing decision-making. Nevertheless, its successful implementation requires addressing critical challenges such as resistance to change, ethical concerns and legal risks. Organizations should focus on fostering human-AI collaboration to unlock the full potential of AI-driven HRM.

Background: Blockchain technologies have come a long way, and integration of blockchain technologies into different fields is flourishing; however, there is a lack of blockchain platforms to manage the high network loads and more sophisticated security threats. These limitations impede the mass adoption of blockchain applications. One of the main reasons blockchain needs artificial intelligence (AI) is to integrate it for the widespread adoption of blockchain technology, as AI addresses scalability and security problems. Objective: The article proposes a pattern analysis model to overcome scalability and security limitations in blockchain systems by applying advanced AI techniques. Methods: To make the model scalable, the proposed model uses deep learning methods such as recurrent neural networks (RNNs) and long short-term memory (LSTM) networks. Furthermore, random forest and convolutional neural networks (CNNs) are applied to augment security operations as an effective classier and anomaly detector on transaction data and a real-time threat detection on transaction patterns using the CNNs. By analysing time series data and dealing with long-term dependencies, the model uses RNNs and LSTMs to enable the strategic introduction of the model to predict and control network loads. Results: When the proposed model is tested against a curated cloud dataset, it significantly outperforms the state-of-the-art approach in all the performance parameters. More specifically, it has exhibited a 5.05% increase in processing speed, 8.05% improvement in energy efficiency, and 5.27%, 5.8%, 10.24% and 11.62% better attack analysis precision, accuracy, recall and AUC, respectively. Conclusion: The synergistic interaction of the applied AI techniques results in a blockchain paradigm that is both scalable and resilient to new security threats. This significant improvement in performance parameters demonstrates the effectiveness of integrating AI with blockchain technology to overcome scalability and security limitations, thereby enabling the widespread adoption of blockchain applications.

Background: Osteoporosis is a condition characterized by bones that are porous and brittle, increasing the risk of fractures. It is often asymptomatic until substantial harm develops, making it crucial to treat at the onset of the disorder.Objective: This study aims to develop a practical, in-depth and adaptable framework utilizing constructed clinical and demographic datasets for the early detection of osteoporosis.Methods: We design a cascade convolutional neural network with adaptive weight fusion and fine-tune it with a real-coded genetic algorithm. An anonymized clinical and demographic record publicly available bone mineral density dataset was used. Missing data identification, normalization and encoding of categorical variables were key diagnostic steps. The training subset consisted of 70% of the dataset, while the remaining 30% was used for testing.Results: The predictive capability of the proposed model is demonstrated by utilizing two datasets. Dataset 1 is used for training and testing, achieving a classification accuracy of 99.5%, precision of 98.7%, recall of 99.0% and AUC-ROC of 0.99. Dataset 2 is used to test the model generalizability, achieving a classification accuracy of 97.0%.Conclusion: The model integrates well into primary care settings, as it relies on structured clinical data rather than imaging. Its low costs relative to value and high scalability make it suitable for population-level screening and treatment of osteoporosis. The limitation of the research is that we utilize only a clinical dataset to train the model without image analysis.

Against the backdrop of increasing smartwatch usage and the dynamic landscape of evolving features, a nuanced understanding of consumer opinions and preferences is vital for tailoring features and crafting effective marketing strategies. This study addresses this imperative by conducting a comprehensive analysis of customer reviews on smartwatches, aiming to determine the pivotal factors guiding consumer purchasing decisions. By employing word clouds to visually represent sentiments, the study uncovers notable trends. Positive reviews prominently highlight the term “quality”, suggesting a strong emphasis on product excellence. In contrast, negative reviews were characterized by the prevalence of the term “fake”, indicating concerns related to authenticity. Additionally, a comparative assessment of two machine learning algorithms, namely support vector machines and Naive Bayes, demonstrates that support vector machines exhibit superior accuracy in classification. These findings offer valuable insights for industry practitioners navigating the competitive landscape of the smartwatch market, providing actionable information for optimizing product features and refining marketing strategies to meet consumer expectations.

The integration of generative artificial intelligence (AI) into adaptive and personalized learning represents a transformative shift in the educational landscape. This research paper investigates the impact of incorporating generative AI into adaptive and personalized learning environments, with a focus on tracing the evolution from conventional artificial intelligence methods to generative AI and identifying its diverse applications in education. The study begins with a comprehensive review of the evolution of generative AI models and frameworks. A framework of selection criteria is established to curate case studies showcasing the applications of generative AI in education. These case studies are analysed to elucidate the benefits and challenges associated with integrating generative AI into adaptive learning frameworks. Through an in-depth analysis of selected case studies, the study reveals tangible benefits derived from generative AI integration, including increased student engagement, improved test scores and accelerated skill development. Ethical, technical and pedagogical challenges related to generative AI integration are identified, emphasizing the need for careful consideration and collaborative efforts between educators and technologists. The findings underscore the transformative potential of generative AI in revolutionizing education. By addressing ethical concerns, navigating technical challenges and embracing human-centric approaches, educators and technologists can collaboratively harness the power of generative AI to create innovative and inclusive learning environments. Additionally, the study highlights the transition from Education 4.0 to Education 5.0, emphasizing the importance of social-emotional learning and human connection alongside personalization in shaping the future of education.

Generative artificial intelligence (GAI) presents an opportunity to democratize access to high-performance, easy-to-use tools of productivity enhancement. However, current adoption patterns suggest that it may instead amplify existing digital divides. The aim of this paper is to propose a policy intervention to ensure equitable access to frontier GAI capabilities: the universal basic AI access (UBAI). Relying on literature research and theoretical analysis, we examine two implementation variants: a voucher-based system making use of commercial providers (UBAI-Light) and direct public provision of GAI services (UBAI-Heavy). We also consider a gradual implementation approach that allows policymakers to support an immediate capture of democratizing benefits while building the capacity for a more substantive future government involvement, should it become necessary. Given the rapid pace of GAI development and adoption, we conclude that timely implementation of UBAI could help prevent the spread of GAI-driven inequalities before they become entrenched.

Background: Static word embedding models such as Word2Vec and GloVe remain widely used in natural language processing, yet key hyperparameters are often selected heuristically rather than through systematic validation.Objective: This study provides an extrinsic evaluation of context window size and embedding dimensionality for Word2Vec (CBOW and Skip-gram) and GloVe embeddings in a downstream spam classification task.Methods: Embeddings were trained on a large external corpus and evaluated using a neural network and several classical machine learning classifiers.Results: The results show that context window size has a moderate influence on performance, whereas embedding dimensionality has a clearer effect: values below approximately 50 degrade performance, while increases beyond moderate ranges (approximately 100–150) yield diminishing returns. Across all experiments, Word2Vec achieves higher stability and performance than GloVe.Conclusion: Overall, the findings suggest that robust classification performance can be achieved with moderate embedding dimensionalities and smaller context windows, providing practical guidance for efficient embedding configuration.

Background: The k-nearest neighbours (k-NN) is a well-established classifier in machine learning. Yet, its performance drops and computational costs rise with extensive or redundant datasets. Furthermore, current instance selection (IS) approaches often face scalability problems and are sensitive to parameter settings.Objective: This study seeks to design a straightforward and efficient IS algorithm that reduces both dataset size and computational demands while preserving or enhancing the accuracy of k-NN classification.Methods: We propose Euclidean ranking-based instance selection (ERbIS), a novel IS approach that prioritises samples based on their Euclidean distance from a single anchor point. In this study, two anchor points are introduced: the first data anchor point (FD) and the mean of each column anchor point (MEC). Both ERbIS models (ERbIS-FD and ERbIS-MEC) are evaluated across 21 KEEL datasets. For performance comparison, the ERbIS models are benchmarked against current and state-of-the-art methods, including condensed nearest neighbour rule (CNN), edited nearest neighbour rule (ENN), adaptive threshold-based instance selection algorithm (ATISA1), decremental reduction optimization procedure (DROP3) and ranking-based instance selection (RIS1). The evaluation focuses on reduction speed, reduction rate and k-NN classification accuracy.Results: The ERbIS models reduce dataset size by an average of 35 to 40% without compromising accuracy compared to the original k-NN and state-of-the-art IS models. Both ERbIS models also demonstrate superior computational efficiency in the reduction process relative to ENN and CNN. Notably, the ERbIS-MEC variant, which utilises the mean of each column as the anchor point, achieves the highest generalisation accuracy among all current and state-of-the-art models.Conclusion: ERbIS offers an efficient and scalable approach for instance selection in k-NN classification, achieving significant data reduction and enhanced predictive accuracy with minimal parameter tuning. The model demonstrates strong potential for application to large datasets and may be further improved by investigating alternative distance metrics or integrating hybrid instance selection strategies.

Background: Recommender Systems (RSs) frequently face challenges in balancing exploration and exploitation, particularly in dynamic environments where user behaviors evolve over time. Traditional methods struggle to adapt to these complexities, limiting their effectiveness in real-world domains such as e-commerce, streaming services, and social networks. Objective: The objective of this study is to introduce DAC-GCN, a Dual Actor-Critic Graph Convolutional Network, designed to enhance recommendation accuracy, ranking quality, and adaptability to evolving user preferences. DAC-GCN merges graph-based learning with Deep Reinforcement Learning (DRL) techniques to improve both short-term and long-term user-item interactions. Methods: DAC-GCN utilizes a dual architecture with separate Graph Convolutional Networks (GCNs) for policy optimization and value estimation. It incorporates Multi-Hop Aggregation (MHA) to capture extended user-item dependencies and an attention mechanism to emphasize pivotal relationships. We evaluate DAC-GCN on benchmark datasets, including MovieLens 100K, MovieLens 1M, Amazon Subscription Boxes, Amazon Magazine Subscriptions, and Mod Cloth, using standard ranking metrics (Precision@K, Recall@K, NDCG@K, MRR@K, and Hit@K). Results: Experimental results demonstrate that DAC-GCN consistently outperforms state-of-the-art baselines, showing significant improvements in recommendation accuracy, ranking quality, and robustness to shifting user behaviors. The model’s ability to capture complex user-item interactions is greatly enhanced by MHA and attention mechanisms, while the dual architecture ensures training stability. Conclusion: DAC-GCN offers a scalable, high-performance solution for modern recommender systems, effectively addressing challenges such as data sparsity and changing user preferences. By integrating graph-based methods with DRL, this study advances both the theory and practice of recommender systems and provides valuable insights for future research and practical applications.

Background: Many real-world tabular datasets are heterogeneous, with distinct regions of the feature space exhibiting different feature–label relationships. Conventional global classifiers often miss these local patterns, reducing both predictive accuracy and interpretability. Objective: This study aims to design a modular classification framework that combines local specialization with global consistency to enhance predictive performance and interpretability in heterogeneous tabular data.Methods: The author proposes Cluster-guided local feature selection with top-2 voting and fallback (CGLFS+), which integrates unsupervised clustering, cluster-specific feature selection and lightweight local models. Final predictions combine top-2 local decisions with a global fallback classifier for robustness. The framework was evaluated on five diverse benchmark datasets using repeated stratified cross-validation.Results: CGLFS+ achieved consistent gains in accuracy and macro F1 over strong baselines, with statistically significant improvements and competitive inference times.Conclusion: CGLFS+ successfully balances local adaptation and global consistency, providing a scalable and interpretable approach well suited to heterogeneous domains such as healthcare, chemistry and finance.

The paper is designed to present the development of scientific research into dynamic pricing in the e-commerce industry. Researchers all over the world attempt to investigate the influence of dynamic pricing on revenue and operational costs and propose models to solve operational or strategic problems in different industries, including e-commerce. In order to understand the level of development of dynamic pricing in e-commerce, a bibliometric analysis is performed. The analysis covers 153 papers collected from the Web of Science database. The analysis reveals that while dynamic pricing research is widespread, it is often explored in industries beyond e-commerce. Key industries, including electricity, airlines, home delivery, transportation and hospitality, have also adopted dynamic pricing strategies. The study highlights the growth of interest in dynamic pricing in e-commerce since 2001, with a peak in 2021. The results represent the growing interest among researchers in dynamic pricing in e-commerce in line with the development of e-commerce as an industry. The analysis underscores the ongoing relevance and expanding scope of research in this domain, with many unexplored aspects and challenges in the dynamic pricing strategies employed by e-commerce businesses.

Background: The preservation and analysis of spoken data in oral history archives, such as Holocaust testimonies, provide a vast and complex knowledge source. These archives pose unique challenges and opportunities for computational methods, particularly in self-supervised learning and information retrieval.Objective: This study explores the application of state-of-the-art artificial intelligence (AI) models, particularly transformer-based architectures, to enhance navigation and engagement with large-scale oral history testimonies. The goal is to improve accessibility while preserving the authenticity and integrity of historical records.Methods: We developed an asking questions framework utilizing a fine-tuned T5 model to generate contextually relevant questions from interview transcripts. To ensure semantic coherence, we introduced a semantic continuity model based on a BERT-like architecture trained with contrastive loss.Results: The system successfully generated contextually relevant questions from oral history testimonies, enhancing user navigation and engagement. Filtering techniques improved question quality by retaining only semantically coherent outputs, ensuring alignment with the testimony content. The approach demonstrated effectiveness in handling spontaneous, unstructured speech, with a significant improvement in question relevance compared to models trained on structured text. Applied to real-world interview transcripts, the framework balanced enrichment of user experience with preservation of historical authenticity.Conclusion: By integrating generative AI models with robust retrieval techniques, we enhance the accessibility of oral history archives while maintaining their historical integrity. This research demonstrates how AI-driven approaches can facilitate interactive exploration of vast spoken data repositories, benefiting researchers, historians and the general public.

Background: Despite its increasing importance, cloud computing is vulnerable to Distributed Denial of Service (DDoS) attacks, affecting data centre availability and functionality. Unfortunately, the impact of these attacks on cloud survivability remains underexplored. Most works overlook long-term resilience and lack comprehensive metrics, in-depth simulation, large-scale experiments, and combined attack and defence scope. Objective: This study investigates the survivability of cloud environments under DDoS attacks in extreme cases, involving intensive attacks leading to cloud failure. By simulating worst-case scenarios, including thousands of attacks on large-scale clouds with and without countermeasures, we assess cloud resilience and identify the limitations of existing defences. Methods: We conduct extensive simulations using NetLogo, modelling a cloud environment subjected to SYN flood, smurf, UDP flood, HTTP flood and malformed packet attacks. We evaluated the impact of attacks individually and in combinations, both with and without countermeasures. Each simulation involves request exchanges between end user nodes and data centres using an appropriate algorithm. We varied parameters like the number of data centres, malicious nodes, and the types and rate of attacks. Results: The study analyses cloud resilience in terms of message delivery, available data centres, and functional node ratios, as well as tolerance and breakage thresholds. Findings indicate that cloud systems can tolerate a certain level of DDoS attack density where data centres remain accessible even without countermeasures. However, the latter greatly enhances cloud security, although their performance may decrease dramatically under extreme conditions. This highlights the importance of optimizing countermeasures, especially to handle high-intensity attacks. Conclusion: This study provides valuable insights for cloud managers to enhance resilience and face sophisticated DDoS attacks. While current countermeasures offer initial mitigation, they are insufficient against complex and combined threats. Thus, future research should focus on developing robust, multi-layered defence mechanisms and providing data centre duplication to ensure service availability.

Background: As assaults get more sophisticated, honeypots like Dionaea become an essential tool for analysing attack behaviours and detecting weaknesses. Despite their growing importance in cybersecurity, honeypots' role in real-time cyberattack surveillance and threat intelligence is largely unknown. Many studies concentrate on identifying attacks rather than delivering actionable intelligence for defensive solutions. Furthermore, previous research frequently lacks thorough methodology for comparing attack data to real-world incidents and does not investigate the integration of honeypots with external intelligence services.Objective: This study assesses the Dionaea honeypot's ability to detect and analyse cyberattack trends, with an emphasis on attack patterns, malware dispersion, and geographical threat sources. The project will look into how Dionaea honeypots, when combined with external analysis services such as VirusTotal, might provide more thorough insights into cyberattack tactics and improve proactive cybersecurity defence mechanisms.Methods: The Dionaea honeypot was used to identify a range of attacks on vulnerable services including Telnet (Port 23), SMB (Port 445), and MySQL (Port 3306). Over a seven-day observation period, 32,395 attack connections from 6,276 distinct IP addresses were detected, yielding 2,892 malware samples. These samples were examined using VirusTotal, and the findings were categorised by malware type, attack vector, and geographical origin. Geospatial and service-specific attack patterns were also investigated to detect emerging trends and high-risk sites.Results: The investigation identified WannaCry ransomware as the most common malware, accounting for 1,076 incidents, demonstrating the continuous exploitation of the MS17-010 vulnerability in SMB (Port 445). The most frequently attacked ports were Port 23 (Telnet), Port 445 (SMB), and Port 3306 (MySQL), which received 7,988, 6,898, and 3,589 attack attempts, respectively. Geographically, the leading sources of assault activity were China (42%), the United States (17%), and Japan (13%). The findings demonstrate that honeypots are not only effective attack detection tools, but also significant sources of intelligence for understanding cyber threat methods and adversary behaviours.Conclusion: This study proposes DORA (Dionaea Observation and Data Collection Analysis), an integrated system that enhances the existing Dionaea honeypot by combining its data with external analysis services like VirusTotal. This integration provides critical insights into real-time cyberattack detection, malware analysis, and attack vector identification. The findings highlight vulnerabilities in services like Telnet and SMB, particularly the exploitation of MS17-010. DORA improves threat intelligence workflows, enhancing malware detection accuracy and classifying threats more efficiently. Additionally, it helps identify high-risk attack surfaces, forming the basis for adaptive cybersecurity strategies. This research contributes to developing resilient defence systems capable of addressing emerging threats.

The literature review was conducted systematically, following a rigorous process to address specific research questions. The review procedure was designed to provide guidance and minimize researcher bias. It outlined the study selection process, including inclusion and exclusion criteria, research questions, search methods, quality evaluation, and data extraction and synthesis. The Scopus database was utilized for this systematic literature review, and a comprehensive search was conducted to identify relevant studies. We used the Kitchenham systematic literature review (SLR) method required to process metadata at the time of processing this SLR, and PRISMA guidelines for reporting systematic literature reviews and meta-analyses. Additionally, VOSviewer analysis was employed to gather data on the sources used by individuals and organizations to access information about fintech products and services, and to understand their influence on acceptance behaviour. A total of 850 publications were identified and screened, with 70 fintech customer acceptance studies meeting the inclusion and exclusion criteria. These studies were published between 2012 and 2022 and were limited to Scopus indexed journals. To maintain focus, specific research questions (RQ) were developed, and data were gathered accordingly to address each RQ while adhering to quality standards. Reviews and quality checklists were used to extract relevant data, prioritizing the most comprehensive publication when multiple sources reported the same data. The primary studies analysed indicated that research into fintech acceptability spans various scientific disciplines, including computer science, information technology, business management and marketing. The technology acceptance model (TAM) emerged as the most used approach for measuring user acceptance of fintech services, as identified in 43 out of 70 publications. Furthermore, several researchers have incorporated additional factors such as performance, social influence, cultural and religious values, knowledge, and service quality to enhance the understanding of fintech acceptance.

Background: Despite the rapid development of Smart Technologies for Independent Living at Home (STILH) among older adults, their market is still underdeveloped. Low awareness is one of the key reasons for the slow uptake of STILH in later life. A significant gap exists in the literature regarding the factors that shape older adults′ awareness of STILH. Objective: The aim is to provide a conceptual overview and empirical test of awareness factors of STILH in later life. An explanatory model is proposed by integrating insights from consumer behaviour, information processing and technology adoption models. Methods: The model is tested with structural equation modelling based on survey data from a sample of 1200 internet users aged 55+ in June 2024. Results: The results support all the proposed hypotheses, indicating that exposure to information about STILH, source expertise, self-source congruity and individuals’ innovativeness directly influence awareness of STILH. Moreover, inherent novelty seeking and self-efficacy influence individuals’ innovativeness, and together with self-source congruity and source expertise positively affect exposure to information about STILH. Conclusion: Based on the study results, interventions can be tailored to help scale up STILH more efficiently, ultimately improving quality of life.

Background: The advancement of big data analytics calls for careful selection of processing frameworks to optimize machine learning effectiveness. Choosing the appropriate framework can significantly influence the speed and accuracy of data analysis, ultimately leading to more informed decision making. In adapting to this changing landscape, businesses should focus on factors such as how well a system scales, how easily it can be used and how effectively it integrates with their existing tools. The effectiveness of these frameworks plays a crucial role in determining data processing speed, model training efficiency and predictive accuracy. As data become increasingly large, diverse and fast-moving, conventional processing systems often fall short of the performance required for modern analytics.Objective: This research seeks to thoroughly assess the performance of two prominent big data processing frameworks—Apache Spark (in-memory computing) and MapReduce (disk-based computing)—with a focus on applying random forest algorithms to predict stock prices. The primary objective is to assess and compare their effectiveness in handling large-scale financial datasets, focusing on key aspects such as predictive accuracy, processing speed and scalability.Methods: The investigation uses the MapReduce methodology and Apache Spark independently to analyse a substantial stock price dataset and to train a random forest regressor. Mean squared error (MSE) and root mean square error (RMSE) were employed to assess the primary performance indicators of the models, while mean absolute error (MAE) and the R-squared value were used to evaluate the goodness of fit of the models.Results: The RMSE, MAE and MSE obtained for the Spark-based implementation were lower, compared to the MapReduce-based implementation, although these low values indicate high prediction accuracy. It also had a big impact on the time it took to train and run models because of its optimized in-memory processing. As opposed to this, the MapReduce approach had higher latency and lower accuracy, reflecting its disk-based constraints and reduced efficiency for iterative machine learning tasks.Conclusion: The conclusion supports the fact that Spark is the better option for complex machine learning tasks such as stock price prediction, as it is good for handling large amounts of data. MapReduce is still a reliable framework but not fast enough to process and not lightweight enough for analytics that are too rapid and iterative. The outcomes of this study are helpful for data scientists and financial analysts to choose the most appropriate framework for big data machine learning applications.

The article aims to explore the role of information resources provided by university libraries in strengthening educational and professional programmes in the field of Information, Library and Archival Studies based on the Scientific Library of Kyiv National University of Culture and Arts. The purpose of this study is to investigate how these resources can contribute to the overall growth and development of students and professionals in the field. Using a descriptive and analytical research methodology, the study examines the diverse range of information resources available in the library, including digital databases, online journals, e-books and other relevant materials. The findings reveal that these resources serve as a solid foundation for enhancing knowledge, skills and competencies required in the field. The practical implications of this research emphasize the importance of utilizing the rich information resources of university libraries to design and implement effective educational and professional programmes. By utilizing these resources, educational institutions and professionals can strive for continuous improvement, staying updated with the latest trends and advancements in the field. This study highlights the critical role of university library information resources in augmenting educational and professional programmes in Information, Library and Archival Studies. The findings underscore the need for collaboration and strategic utilization of these resources to shape well-rounded professionals capable of meeting the evolving demands of the information age.