Introduction. Every day, the number of news, pages on social networks and chats on the Internet is increasing, accordingly, there is an increase in information that carries an emotional load. At the same time, the number of information threats is also growing. Under such conditions, the construction of systems for determining the emotional color of texts becomes extremely relevant. Purpose. Emotional messages can be found and classified using artificial intelligence, namely based on neural network methods. For the process of learning neural networks, it is necessary to have a training sample of texts with a preliminary assessment of their emotional coloring. Such marked learning samples exist for news and texts in English, however, at the moment, no accessible learning sample of Ukrainian news and texts has been created. Methods. Using statistical methods of sentiment analysis for detecting text tonality with extended vocabulary. Results. Extended tonality vocabulary of the Ukrainian language was built. A large corpus of texts and their emotional coloring was built with an expertly assessed markup accuracy of 98%, containing 5,318,783 texts of various types in the Ukrainian language. Conclusion. The built text corpus can be used to train and test neural networks for sentiment analysis of Ukrainian-language texts.
Introduction. Diffuse diseases are the most numerous class of liver diseases. Among them, autoimmune hepatitis stands out for its severe course in children. Its timely diagnosis and assessment of the degree of liver damage is an integral part of a patient’s personalised treatment strategy. The lack of reliable non-invasive methods for assessing liver disease affects the quality of medical services. Therefore, the search for informative signs of liver damage in ultrasound images and the improvement of methods for solving multi-class classification problems are relevant areas for the development of non-invasive systems for determining the degree of liver fibrosis. Purpose. Improve the diagnosis of liver fibrosis stages through a multi-level classification system. Methods. A system for classifying the detailed degree of fibrosis (eight classes) based on neural networks according to the state of the blood vessels in ultrasound images of the liver is proposed and substantiated: the first level is a fibrosis degrees group classification of fibrosis degree for regions of interest by convolutional neural networks, the second level is the classification of fibrosis individual degrees for regions of interest by a deep neural network, the third level is the integration of the second level results to obtain conclusions about the patient (image) as a whole. In order to optimize the feature space, we have performed an exploratory analysis using a logistic multivariate regression model optimized by the Group Method of Data Handling. The resulting set of generalized variables formed the meta-feature space for the second level of the system. A twofold increase in the quality of the system’s classification is shown in comparison with solving the task of image classification by a single convolutional network with an output of eight classes. Results. Improved version of the hierarchical system for solving multiclass problems based on the use of ANNs is proposed. The system implements the classification of the detailed degree of liver fibrosis in children with autoimmune hepatitis using ultrasound images characterizing the state of liver vessels. The use of a hierarchical classification system allowed us to obtain a classification accuracy of 32.61% higher than the use of a standard multi-class classifier based on a convolutional neural network. The classification accuracy of the hierarchical system: at the first level – 32.46%; at the second level – 50.43%; at the third level – 65.22%. Conclusion. The article proposes, substantiates and develops a hierarchical classification system based on convolutional neural networks. Its use makes it possible to increase the accuracy of classification of the detailed degree of liver fibrosis by 2 times compared to the standard multi-class classifier based on СNNs. The main source of further improvement of the classification accuracy of the system should be a combination of signs of vascular deformation and texture features that can be obtained with different ultrasound imaging modes. The developed system offers new opportunities for improving methods for solving multiclass classification problems based on image analysis.
The health status of patients is recorded in various sources, such as medical records, portable devices (smart watches, fitness trackers, etc.), forming a characteristic current health status of patients. The goal of the study was the development of medical card software for the analysis of data from fitness bracelets. This will provide an opportunity to collect data for further use of cluster analysis and improvement of the functionality and accuracy of medical monitoring. The object of the study is the use of linear regression to analyze and predict heart rate based on data collected using fitness bracelets. In order to solve this problem, an information system was developed that uses linear regression to analyze the effect of parameters such as Very Active Distance, Fairly Active Minutes, and Calories on the heart rate (Value). Training and validation were performed on data from fitness bracelets. The results confirm the effectiveness of linear regression in predicting heart rate based on the parameters of fitness bracelets. The accuracy of the model was compared under the conditions of aggregation and without it, which allows us to draw conclusions about the optimal conditions for using linear regression for the analysis of fitness data. The study proves the adequacy of the obtained results according to the Student’s criterion. The calculated Student’s t test is 1.31, with the critical test ¾ 2.62. Which proves the adequacy of the developed model. The results of the study confirm that the linear regression model is an effective tool for individual monitoring and optimization of physical activity based on data from fitness bracelets. It is worth considering that the use of linear regression has its limitations and is not always the best choice for complex nonlinear dependencies. In such cases, other machine learning methods may need to be considered.
Introduction. Interest in the study of digital platforms (DP) is due to their prevalence and the dependence of this phenomenon on the possibilities of using information technologies. The growing distribution and great potential of the DP is connected not only with the use of new hardware and software, but also with the integration of digital technologies into business processes. The need for a deeper understanding of the differences and similarities of various CPUs prompts researchers to turn to the fundamental mechanism of knowledge organization – classification. From a practical point of view, the classification helps to compare different CPUs and allows users to choose the one that provides the desired results. Formulation of the problem. The problem of CPUs classification is to identify specific and common characteristics for building clusters using different approaches. When modeling and solving the classification problem, static methods and machine learning methods are used. The most widespread of them are the method of nearest neighbors and the method of support vectors. The theory of combinatorial optimization was used to build the mathematical model. The approach proposed. To build a mathematical model of the classification problem, the theory of combinatorial optimization was used, which allows to investigate some properties of this problem. The argument of the objective function in it is the division of the -element set into subsets. This combinatorial configuration can be either with or without repetitions, either finite or infinite. When finding the optimal result, a situation of uncertainty arises, which is related to the structure of the argument of the objective function which is a combinatorial configuration. Conclusion. The classification problem belongs to a broad class of partitioning problems. In it, the characteristics of the clusters are known, the objects that need to be determined, to which class they belong, are analyzed not simultaneously, but by groups or individual elements. Since the result is determined not simultaneously, but by a partial objective function, the classification problem belongs to the dynamic problems of combinatorial optimization. The classification of digital platforms is carried out by heuristic methods, in particular the nearest neighbor method. Both one and a set of common characteristics characteristic of certain CPUs are used as criteria.
Introduction. The modeling of a complex object “data analysis of learning of the subject throughout life”, supported by technology, is experiencing a special stage of its development, undergoing a great influx of potential opportunities and possibilities. induce a steady increase in digital capabilities for everyone, Numerous subjects implement the designated capabilities with different perspectives, goals, at different levels, stages, different approaches, methods, designs, languages, procedures, systems, processes, tools, services, standards The hidden problem It seems that this great potential has not yet been systematically realized throughout life. And therefore, a lot of existing knowledge, models and technologies are often not effectively translated into existing tools for everyone. In our research, modeling focuses at a high level of abstraction on the enhanced understanding of the subject of the strategy for direct development, the adoption of informed solutions to the selection, adaptation of existing and planned Innovative tools, methods, analytics of all types with the help of available management systems. Purpose. The purpose of this study is develop a formalized description with meaningful interpretations of basic system-forming elements, modeling constructs, a general model, inheritance models and a register of tasks to systematically improve understanding, progress of results, quality of products, services and making informed decisions for stakeholders based on methods and tools data analysis of learning of the subject throughout life. Methods. System methodology, methods of analogies, didactic methods. Results. On the basis of fundamental facts, ideas and systematic methodology, at the highest level of formalization, basic system elements, modeling constructs, a general model, inheritance models and a register of tasks are proposed and meaningfully interpreted in order to systematically improve understanding, progress, results, quality of products, services and acceptance reasoned decisions for interested parties based on methods and tools of of learning of the subject throughout life with the help of an accessible management system. Conclusion. Modeling and practical implementation of an extremely complex process, system in the era of digital transformations requires a comprehensive solution to many complex problems such as understanding, scaling, protection of property, elimination of uncertainty, interoperability, harmonization of existing and planned official and de facto standards. Systematized application of constructions from mathematical theories allows to better see their behavior, destroys uncertainty, helps to scale solutions, etc. Therefore, a necessary condition, a requirement for systematic improvement of models is a complex interpretation of abstractions in the context of the specified problems, as well as their practical approbation using available control systems with the aim of identifying and disseminating best practices to interested parties. The main directions of further research: building models of learning oriented games as part of the developed model of the Register of tasks in order to improve the skills of subjects in relation to data analysis: such as critical thinking, problem solving, communication, subject knowledge, data visualization; research on best practices for using the Glossary.
The concept of analysis is being contemplated in its most comprehensive philosophical context. An endeavour is undertaken to systematise the examination and present it in the form of a systematic procedure, in essence, to construct a formal comprehensive analysis model. The generalised model combines normal data processing procedures with pairs of philosophical categories of the most generic form as components. The sequence of their application is determined. This order is based on the degree of generality of the categories. Consequently, several analysis models were acquired. Despite the observable consistency in the sequence of category application, a comprehensive analytical model has not yet been established based on this series. However, the analysis techniques that have been obtained can already be regarded as prototypes of practical algorithms, serving as the foundation for deductively deriving algorithms for actual computer programmes.
Introduction. The necessity for modern approaches to solving speech recognition tasks arises from the rapid development of artificial intelligence and the need to improve the accuracy and speed of human-computer interaction in various areas, such as voice assistants, translation, and automation. This direction is becoming increasingly relevant due to the growing volume of generated audio data and the need for real-time processing, particularly in Ukrainian contexts where multiple languages and dialects coexist. Currently, several approaches to speech recognition, analysis, and transcription exist, including methods based on neural networks, speaker diarization techniques, noise removal, and data structuring. However, the challenge of creating a universal solution that meets the needs of multilingual environments and effectively handles unstructured audio data remains relevant. Purpose. To review existing tools and algorithms for solving speech recognition tasks, particularly for Ukranian. Methods. Speech recognition, deep learning, transformers. Results. Theoretical foundations of approaches and models for speech recognition were considered for building a knowledge base for a multilingual spoken dialogue system. Effective examples of improving transcription accuracy for languages with limited data were also explored, along with potential steps to enhance system speed. Potential datasets for model training were discussed. Conclusion. A structured review of modern methods for processing and analyzing multilingual audio files was provided, outlining their advantages, disadvantages, and unresolved issues.
Introduction. In recent years, there has been significant growth in the use of Generative Artificial Intelligence (Generative AI) applications that create visual data based on textual descriptions. This opens up new possibilities for enhancing information systems across various applied fields. It goes beyond just vivid forms of representation or information display; it offers the potential to address complex practical tasks at a fundamentally different level. The ability to store the history of visualizations allows tracking the dynamics of changes in specific information over time, providing deeper information analysis and a higher level of decision-making and strategy formulation in complex business systems. However, the semantic quality of generating visual content remains a challenge, influenced not only by the choice of the generation model itself but also by the accuracy of the input instructions. This article is dedicated to analyzing the current state of Generative AI technologies, existing models, their functional capabilities, advantages, limitations, and drawbacks. Purpose. The research aims, on one hand, to identify effective applications of these technologies in solving complex business tasks, and on the other hand, to highlight issues with existing models regarding their integration into the business process chain and possible ways to address them. Results and conclusion. The analysis conducted allowed: Identifying common features of existing systems and highlighting characteristics significant for choosing a model in terms of its integration into a complex system, such as the semantic accuracy of generation results, openness of code, price, multilinguality, etc. Formulating directions for further research, primarily focused on developing methodological foundations for automatically generating input textual descriptions for Generative AI models in video content generation.
Introduction. Object recognition on aerial images is an urgent task in modern conditions, especially in cases requiring accurate and fast car recognition. Traditional contour extraction methods, such as Canny, Sobel, Laplacian, Prewitt, and Scharr, are based on gradient analysis and are known for their ease of implementation. This is an essential step for further recognition, as the correct definition of contours contributes to more accurate object identification. However, the effectiveness of the above methods could be improved, especially in complex environments with high object density, uneven brightness, and noise. Neural network models, such as YOLO (You Only Look Once), offer new possibilities, providing more accurate and reliable recognition, even in difficult situations. Purpose. This study compares the effectiveness of classical contour extraction methods and the YOLOv6n neural network model for vehicle recognition in aerial images. The accuracy of vehicle detection is evaluated by the main metrics: Precision, Recall, and F1-measure, which allow the determination of each method’s efficiency level in specific conditions. Methods. The study includes testing the classical Canny, Sobel, Laplacian, Prewitt, and Scharr algorithms for car outline detection and analyzing the results of the YOLOv6n model for deep-learning object detection. Classical methods use image processing to identify contours based on pixel gradients, which allows for extracting structures in an image. The YOLOv6n model is based on a neural network approach, considering complex image features for more accurate and faster object detection. Results. The data analysis showed that classical methods, although they can detect contours, have limited accuracy in conditions of high object density and sharp changes in brightness. The accuracy (Precision) and F1 Score for traditional methods was low, indicating a significant number of false positives and false negatives. In particular, the Sobel and Scharr methods showed the highest Recall but significantly decreased accuracy. In contrast, the YOLOv6n neural network model demonstrated high results in all primary metrics: Precision – 97.9%, Recall – 94.8%, F1 Score – 96.32%, and maP – 97.6%, which confirms its advantages in providing accurate and reliable vehicle recognition in aerial images. Conclusions. The study has shown that traditional contour extraction methods can serve as auxiliary tools for image preprocessing. Still, they need to provide adequate accuracy for the final stages of vehicle recognition. Neural network approaches, such as YOLOv6n, significantly outperform classical methods by providing high detection speed and accuracy, making them recommended for use in high-precision object recognition tasks in aerial images.
Introduction. Despite the rapid development of the chemical industry and science, discoveries in the field of health care, the emergence of drugs and therapeutics based on nanotechnology and the development of radiation therapy technologies, the safety of biomedical applications of the latest products, and the search for new methods and approaches to the diagnosis and treatment of cancer are an open issue. One of the safest and fastest methods for researching the behaviour of new materials and tools and selecting the best candidates is the modelling of relevant processes, particularly computer molecular modelling based on mathematical models. However, despite a large number of available methods and modelling tools, for most of them, the successful application is possible only for a narrow range of tasks and experiments. As one of the possible solutions to this problem, we propose a new approach to computer molecular modelling based on the synergy of the algebraic approach, namely, algebraic modelling and biological knowledge at different levels of abstraction, starting from quantum interactions to interactions of biological systems. We see one of the directions of application of this approach in the possibilities of modelling the radiation therapy process – starting from modelling the accelerators’ work and ending with modelling the interaction of the particles’ beam with the matter at the level of quantum interactions. In particular, in the article, we consider the possibilities of forward (specific and symbolic) and backward (symbolic) algebraic modelling on the example of models of the higher level of abstraction, which allows us to visualize certain interactions and to build charts of dependencies for specific models, and to determine the presence of the desired scenarios (forward modelling) or a set of initial environment parameters (backward modelling) in symbolic form.
This paper examines the performance of search and multi-agent algorithms within the context of the Pac-Man game. The game is used as a platform to simulate autonomous system management tasks, where an agent must complete missions in a two-dimensional space while avoiding dynamic obstacles. Classical search algorithms such as A* and BFS, along with multi-agent approaches like Alpha-Beta, Expectimax, and Monte Carlo Tree Search (MCTS), are analyzed in terms of their effectiveness under different maze complexities and game conditions. The study explores how maze size, ghost behaviors, and environmental dynamics influence the performance of each algorithm, particularly in terms of execution time, score, and win percentage.
Applying polypoint transformation to a triangle mesh is a promising technique that deserves extensive exploration. Unlike traditional deformation techiques used in 3D animation that are based on spatial interpolation, polypoint transformation transforms not a set of points or mesh vertices, but a polyfiber — a set of planes that form an object of transformation. This can be beneficial in practical applications such as computer animation, deformative modeling in CAD, or deformation prediction in additive manufacturing since, with polypoint transformation, the object of deformation includes the topological information about the mesh and not just its vertex positions. The way a polyfiber can form a triangle mesh is, however, an understudied problem. This problem gets easier to study if we start from a 2D case where a structure that corresponds to a triangle mesh in 3D is a polyline or a conotur made of line segments. For this structure we can decompose the study into a set of explorable questions one of them being: how the configuration of contour forming polyfiber affects the results of transformation, or, more specifically, how an angle between two lines forming a vertex affects the translation of that vertex under a polypoint transformation? Would this dependency prohibit the application of polypoint transformation to 3D mesh deformation? This article answers both questions.
Introduction. Environmental pollution has a significant impact on people’s lives. Drinking water pollution with heavy metals is especially noticeable for humans. To solve this problem, it is necessary to ensure continuous monitoring of water quality, which will allow determining the concentration of toxic elements in it. It is necessary to isolate the useful component of the signal containing information on unknown concentrations of the measured elements, against the background of a mixture of various signals of elements present in the background solution. For this purpose, a method for constructing a basic model is proposed, according to which it is possible to separate the differential signal of the inversion of chemical elements in water from the background signal of impurities present in water. Due to this, a spectrum of a multicomponent intensity signal is formed in pure form, the analysis of which allows one to accurately estimate the unknown concentrations of a mixture of these dissolved elements. Purpose. To develop a method for constructing an approximation function for the lower envelope of the background intensity signal in different classes of basic functions using GMDH in the problem of determining the concentrations of chemical elements in multicomponent signals when measuring the ecological state of environmental objects using electrochemical methods of inversion chronopotentiometry. Methods. The methods that are used in this article are of inversion chronopotentiometry method and GMDH neural network. Results. The problem of constructing a baseline for the multicomponent signal of the intensity of inverse chronopotentiometry, the determination of which allows to estimate the concentration of various chemical elements dissolved in water quite accurately, is investigated. To solve this problem, an approach to construction of the approximation function of the lower envelope line of the differential signal in different classes of basic functions with the use of GMDH is proposed. The approach was used for constructing the best model of the differential signal baseline on the real example of measuring the Zn concentration under the presence of ions Cd, Pb, Cu. The built model of optimal complexity is the sum of arguments with the direct and inverse degrees which is necessary for clearing the intensity signal from background to obtain the intensity spectrum of the measured chemical elements. Conclusion. Produced in the C2 class, it can be recommended for use in the task of providing the baseline of a differential signal, since it can practically be the same R2, but also the richly shortening MAPE.
The article is devoted to solving the problem of determining the resilience of critical infrastructure systems to malicious actions of adversaries. Different modeling methods and their integration are considered. Using the example of a system of systems, including energy and transport networks, the application of methods of agent, network, economic modeling and the method of system dynamics are considered, which are combined into a single structure of analysis for the development of algorithms for general decision-making support for the protection of critical infrastructure systems.
Map representation and management for Simultaneous Localization and Mapping (SLAM) systems is at the core of such algorithms. Being able to efficiently construct new KeyFrames (KF), remove redundant ones, constructing covisibility graphs has direct impact on the performance and accuracy of SLAM. In this work we outline the algorithm for maintaining dynamic map and its management for SLAM algorithm based on Gaussian Splatting as the environment representation. Gaussian Splatting allows for high-fidelity photorealistic environment reconstruction using differentiable rasterization and is able to perform in real-time making it a great candidate for map representation in SLAM. Its end-to-end nature and gradient-based optimization significantly simplifies map optimization, camera pose estimation and KeyFrame management.
Introduction. In modern conditions of development of the world economy, the digital economy is one of the most relevant and important factors in ensuring economic growth. The digital transformation of the economy is a primary direction of innovative development of socio-economic systems, and therefore a tool for creating long-term competitive advantages of the transport and logistics system. The article examines the issue of economic problems associated with obtaining an integrated assessment of the level of competitiveness of the transport and logistics system. The purpose of the article is to analyze the impact of the digitalization of the economy on the competitiveness of the transport and logistics system and to create a model for obtaining an integrated evaluation of its competitiveness based on factors that reflect the specifics of the services provided by the system in the conditions of the digital transformation of the economy. Research methods. The digital transformation of the transport and logistics system is carried out based on the implementation of digitalization at all levels of business processes of economic structures: from the optimization of the logistics of physical flows and the optimization of data exchange to customer service. To assess the level of competitiveness of the transport and logistics system, it is proposed to calculate the competitiveness index by group of services in the process of transporting material flow. The model for calculating the integrated indicator of the competitiveness of the transport and logistics system is described, which is a three-stage multifactor model. Results. The economic evaluation of the digital transformation of the development of logistics services in railway transport allows for specifying the specifics of service services and clarifying the possibilities of using logistics methods of service response in carrying out cargo transportation. This makes it necessary to clarify the place and role of service support of cargo flow in the transport and logistics system and points to bottlenecks that must be overcome in the future to increase the level of competitiveness not only of the system but also of the country as a whole. Conclusions. The development of the digital transformation of the transport and logistics system consists of solving various socio-economic and technological tasks, which include the emergence of new products and new markets; reduction of costs for conducting business activities; and increasing productivity and efficiency of logistics and transport processes. As a result is increased competition in the transport and logistics system.
Introduction. The use of functional magnetic resonance imaging (fMRI) allows for the assessment of processes occurring in the brain. By analyzing the examination results, it is possible to establish the parameters of connections between brain structures, and changes in the values of these parameters can be used as diagnostic conclusion predictors for PTSD-patients. Purpose. To identify predictors for the classification of the PTSD diagnosis using the connectivity parameters of BOLD signals from brain structures. Methods. The technology for identifying predictors of PTSD diagnosis is based on a) the formation of connectivity parameters of BOLD signals from brain structures obtained during resting-state scanning, b) the use of classifier-oriented selection based on inter-class variance and mRMR criteria to select informative features, and c) the classification of PTSD diagnosis using a logistic regression algorithm optimized by the Group Method of Data Handling. Results. The technology proposed in this work enabled the selection of informative features and the identification of their predictive forms, resulting in the formation of classifiers for the diagnosis of PTSD with high accuracy, sensitivity, and specificity. Conclusion. A technology for the formation, selection, and use of connectivity parameters of BOLD signals from brain structures has been proposed for differentiating healthy individuals from those who suffer with PTSD. A list of the most informative features of PTSD and their predictive forms in the form of generalized variables has been obtained, which can be used for diagnostic conclusions. The results obtained indicate the presence of a specific type of connection between the brain areas identified in the study based on levels of excitation (parameters а0 of the models) and the alteration of these levels in the context of PTSD.
The exponential growth of electronically stored textual data poses a significant challenge for search engine developers. This paper is dedicated to a detailed study and comparison of three classical full-text search algorithms: Knuth-Morris-Pratt (KMP), Boyer-Moore (BM), and Rabin-Karp (RK). These algorithms are widely used in computer science for efficient substring searching in textual data. The research results allowed us to identify the strengths and weaknesses of each algorithm and to determine the conditions under which each algorithm is most efficient.
Language learning benefits from a comprehensive approach, but traditional software often lacks personalization. This study analyzes prompt engineering principles to implement a test generation algorithm using Large Language Models (LLMs). The approach involved examining these principles, exploring related strategies, and creating a unified prompt structure. A test generation script was developed and integrated into an API for an interactive language learning platform. While LLM integration offers highly effective, personalized learning experiences, issues like response time and content diversity need addressing. Future advancements in LLM technology are expected to resolve these limitations.
In this paper, we present a method to compute the coefficients of a complex exponential polynomial of real argument that, while being decomposed into real and imaginary parts by Euler’s formula, obtains required interpolating and differential properties at any given points of its real graph. Moreover, imaginary components in their nodes of interpolation and differentiation serve as additional control tools that shape the polynomial appearance. Although the impact of these components is not yet studied extensively, we can still use them to achieve useful properties, e. g. we can minimize the total height of the polynomial graph. From the geometry standpoint, having these properties implies that the parametric curves constructed with such polynomials can go through given points, have predetermined tangent vectors in those or other points, and retain enough variability to have additional useful properties, for instance, the total length of these curves, or their maximal curvature can also be minimized within limits.
