Information and Software Technologies

The interest of musicians and computer scientists in AI-based automatic melody harmonization has increased significantly in the last few years. This research area has attracted the attention of both teachers and students of Theory, Analysis and Composition, looking for support tools for the learning process. The main problem is that the systems designed and developed up to now harmonize a melody written by a user without considering the didactic and therefore cognitive aspects at the basis of a “significant learning”: given a melody, the system returns a harmonization finished without any user input. This paper describes a self-learning algorithm capable of harmonizing a musical melody, with the aim of supporting the student during the study of Theory, Analysis and Composition. The algorithm, on the basis of the ascending and descending movement of the sounds of the melody (soprano), proposes the sounds for the bass line: the Viterbi algorithm was applied to evaluate the probability of the best match between the melody sounds and the provided Markov chains, to reach the “optimal” state sequences. Subsequently, the algorithm allows the user to complete the chords for each sound of the bass line (tenor and alto), or to create the complete chords. Examples of musical fragments harmonized in this way demonstrate that the algorithm is able to respect the concatenation rules of the tonal functions which characterize classical tonal music.

In this paper, we investigate the statistical properties of the CTR mode of a previously presented block cipher based on the matrix power function. Relying on the obtained results we propose an improvement of our original idea to achieve a better mixing of bits. We demonstrate that the modified version of our cipher satisfies both the avalanche effect and the bit independence criterion. To evaluate the quality of the obtained results we compare them to the statistical properties of widely used AES-128 and TDES CTR modes of encryption. Additionally, we present the preliminary analysis of collisions for the CTR mode of our cipher.

This paper presents a metaheuristic algorithm-based proportional-integral-derivative (PID) controller tuning method for a 3 degrees of freedom (DoF) robotic manipulator. In particular, the War Strategy Optimisation Algorithm (WSO) is applied as a metaheuristic algorithm for PID tuning of the manipulator, and the performance of the controller is compared with Particle Swarm Optimisation (PSO) and Grey Wolf Optimisation (GWO) algorithms. According to the simulation outcomes, the WSO algorithm exhibits superior performance compared to the other two algorithms with respect to settling time, overshoot, and steady-state error. The proposed technique provides an effective approach for enhancing the performance of robotic manipulators and can be extended to other applications that require optimal PID controller tuning.

Cryptocurrency price prediction and investment is a popular and relevant area of business nowadays. It involves analyzing historical data to forecast future trends and movements in asset prices. Bitcoin has gained significant prominence in the worldwide financial market as an investment asset. However, the high volatility of its price has attracted considerable attention from researchers and investors alike, leading to a growing interest in understanding the factors that drive its movement. This paper builds upon a research and conducts an empirical approach into the time-series data of a diverse range of exogenous and endogenous variables. Specifically, in this paper, the closing prices of Bitcoin, Ethereum and the daily volume of Bitcoin-related tweets are examined. For forecasting closing Bitcoin price based on the above mentioned predictors, bidirectional long-short term memory (BiLSTM) network tuned by hybrid adaptive reptile search algorithm is proposed. The analysis covers a three-year period from January 2020 to August 2022 and employs a three-fold split of the data to train, validation, and testing datasets. The best generated model by algorithm introduced in this manuscript is compared to other BiLSTM networks tuned by other cutting-edge metaheuristics and achieved results revealed that the method introduced in this research outperformed all other competitors regarding standard regression metrics.

Malware poses a significant global cybersecurity challenge, targeting individuals, businesses, institutions, and nations by compromising sensitive information and causing disruptions, incurring substantial costs. Android devices, with relatively lower security measures allowing installations from unknown sources, face notable malware prevalence, creating opportunities for cybercriminals to engage in illicit activities. To address this issue, numerous research studies have focused on harnessing the power of artificial intelligence (AI) to develop effective solutions. Notably, research utilizing the CICMalDroid2020 dataset has achieved promising results by employing Deep Learning and Machine Learning approaches for Android malware detection. However, to the best of our knowledge, no prior studies utilizing this dataset have explored the potential of the Extra-Tree Machine Learning classifier.

In our research, we endeavored to fill this gap by implementing the Extra Tree classifier in conjunction with cross-validation techniques. Additionally, we employed the SelectFrom-Model feature selection method to enhance the accuracy of malware detection. Through our investigation, we found that the ExtraTree classifier exhibited good performances, achieving an accuracy rate of 96.7%.

Financial statements are typical financial distress identification data for the enterprise. However, nowadays, the valuable data source characterizing enterprise could be expanded, including data from legal events, macro, industry, government register center, etc. This data creates valuable information, which could lead to more accurate financial distress classification model creation. On the other hand, the new data source involvement expands the dimensional space of features and increases the data sparsity. In order to reduce dimensions and have maximum information retention from the initial data space is used feature extraction techniques. This study uses an autoencoder as a nonlinear feature extraction method. Moreover, we compared several structure composition strategies for autoencoders: 1) all data compress; 2) union of the several autoencoders (i.e. data compress of each data type separately and the union of these separate autoencoders). After implementing different autoencoder strategies, eight machine-learning models for financial distress classification were used. The results demonstrated that features retrieved from the union data source strategy outperform the features extracted all at once. These findings create a novelty of autoencoder usage as a feature extraction technique for financial distress key feature’s identification and better financial distress issue classification.

This article discusses the importance of effective scope assessment in Agile projects. The adoption of Agile principles and methods has replaced the traditional waterfall approach, leading to better involvement of clients and teams in gathering requirements, regular review of results, and flexible adaptation to changes. However, Agile implementation projects lack the opportunity to pay detailed attention to the analysis phase, which makes scope assessment crucial. The article proposes using Story Map and UML models to analyze the scope of Agile projects, as their ability to visualize information and act as a source of truth. The methodology helps to integrate models, compile functional and non-functional requirements, perform cross-checking of the requirements, and create an initial scope assessment. Proper preparation for scope assessment is necessary for successful Agile implementation and can lead to understanding project goals and objectives, prioritizing work, managing stakeholder expectations, estimating timelines, and enhancing team collaboration.

The World Economic Forum in Davos in 2022 raised the issue of knowledge by describing the situation as follows: “It could be that we are drowning in content, but starved of knowledge and therefore often fail to connect the dots to anticipate change before it becomes mainstream. With over four billion pieces of content being created each day, keeping abreast of all that is happening far exceeds our capacity to do so. The business models of social media organizations and news outlets have been increasingly focused on giving people more of what they like, leading to echo chamber effects and making it easy to lose sight of the big picture [10].” In recent decades a shift to the knowledge society has been acknowledged, characterized by its ability to identify, create, process, transform, disseminate and use information to generate and use knowledge for the development of individuals [2]. In such a society, intellectual capital is considered to be the most important indicator of wealth, ahead of assets. The acquisition, application and creation of knowledge is more important to the knowledge society than the creation and consumption of information. In regards to knowledge society requirements this paper presents a conceptual knowledge discovery framework: User Interaction and Response-based Knowledge Discovery Framework – UIS-KDF. The framework introduces a meta-level approach for knowledge discovery system design principles.

The online sharing of personal health data by individuals has raised privacy concerns. This paper presents a Named Entity Recognition (NER)-based analysis to detect potential privacy risks in German patient forums. The objective is to extract sensitive information from user-generated texts and augment existing digital profiles of users to demonstrate the potential threats posed by the aggregation of information. To achieve this, we trained a NER model on a large corpus of German patient forum texts and evaluated its performance using standard metrics. The results show that the NER model can effectively extract health-related information from German texts with a micro-average precision of 0.8666, a recall of 0.9633 and an F₁-score of 0.9124. This enables the creation of Digital Twins that accurately reflect the health-related characteristics of individuals. However, when this information is combined with data from different platforms, it poses a potential threat to users’ privacy and underlines the need to warn users.

The publication trends and bibliometric analysis of the research landscape on the applications of machine/deep learning in energy storage (MES) research were examined in this study based on published documents in the Elsevier Scopus database between 2012 and 2022. The PRISMA technique employed to identify, screen, and filter related publications on MES research recovered 969 documents comprising articles, conference papers, and reviews published in English. The results showed that the publications count on the topic increased from 3 to 385 (or a 12,733.3% increase) along with citations between 2012 and 2022. The high publications and citations rate was ascribed to the MDLES research impact, co-authorships/collaborations, as well as the source title/journals’ reputation, multidisciplinary nature, and research funding. The top/most prolific researcher, institution, country, and funding body on MDLES research are; is Yan Xu, Tsinghua University, China, and the National Natural Science Foundation of China, respectively. Keywords occurrence analysis revealed three clusters or hotspots based on machine learning, digital storage, and Energy Storage. Further analysis of the research landscape showed that MDLES research is currently and largely focused on the application of machine/deep learning for predicting, operating, and optimising energy storage as well as the design of energy storage materials for renewable energy technologies such as wind, and PV solar. However, future research will presumably include a focus on advanced energy materials development, operational systems monitoring and control as well as techno-economic analysis to address challenges associated with energy efficiency analysis, costing of renewable energy electricity pricing, trading, and revenue prediction.

Controller management platforms are part of the rapidly growing IoT infrastructure. Platforms manage physical devices and collect, process and integrate data, making them an attractive target for cybercriminals. Weak access control is one of the key cybersecurity threats in this area. This paper aims to provide a secure platform for remote control of controllers using a tailored access control approach. It also aims to evaluate the effectiveness of the proposed access control method. The implemented platform is configured for smart home solutions. Experiments on the administrative cost, speed and security of the method are carried out in scenarios.

The rapid growth of digital information and the increasing complexity of user queries have made traditional search methods less effective in the context of business-related websites. This paper presents an innovative approach to improve the search experience across a variety of domains, particularly in the industrial sector, by integrating semantic search and conversational large language models such as GPT-3.5 into a domain-adaptive question-answering framework. Our proposed solution aims at complementing existing keyword-based approaches with the ability to capture entire questions or problems. By using all types of text, such as product manuals, documentation, advertisements, and other documents, all types of questions relevant to a website can be answered. These questions can be simple requests for product or domain knowledge, assistance in using a product, or more complex questions that may be relevant in determining the value of organizations as potential collaborators. We also introduce a mechanism for users to ask follow-up questions and to establish subject-specific communication with the search system. The results of our feasibility study show that the integration of semantic search and GPT-3.5 leads to significant improvements in the search experience, which could then translate into higher user satisfaction when querying the corporate portfolio. This research contributes to the ongoing development of advanced search technologies and has implications for a variety of industries seeking to unlock their hidden value.

When the body’s defense against an infection damages its own tissues and causes organ malfunction, it develops sepsis, a catastrophic medical illness. Administering intravenous fluids and antibiotics promptly can increase the patient’s chances of survival. In order to determine the best treatment plans for septic patients, this study investigates the application of deep reinforcement learning and continuous state-space models. The method produces clinically comprehensible policies that could assist doctors in intensive care in empowering medical professionals to make informed decisions that ultimately enhance the prospects of patient survival.

This paper delves into the challenges encountered in decision-making processes within Hybrid Energy Systems (HES), placing a particular emphasis on the critical aspect of data integration. Decision-making processes in HES are inherently complex due to the diverse range of tasks involved in their management. We argue that to overcome these challenges, it is imperative to possess a comprehensive understanding of the HES architecture and how different processes and interaction layers synergistically operate to achieve the desired outcomes. These decision-making processes encompass a wealth of information and insights pertaining to the operation and performance of HES. Furthermore, these processes encompass systems for planning and management that facilitate decisions by providing a centralized platform for data collection, storage, and analysis. The success of HES largely hinges upon its capacity to receive and integrate various types of information. This includes real-time data on energy demand and supply, weather data, performance data derived from different system components, and historical data, all of which contribute to informed decision-making. The ability to accurately integrate and fuse this diverse range of data sources empowers HES to make intelligent decisions and accurate predictions. Consequently, this data integration capability allows HES to provide a multitude of services to customers. These services include valuable recommendations on demand response strategies, energy usage optimization, energy storage utilization, and much more. By leveraging the integrated data effectively, HES can deliver customized and tailored services to meet the specific needs and preferences of its customers.

Financial processes are complex procedures related to financial data recording and analysis. Compliance of these processes with the normative rules is important because it is related to the correctness of financial data records, it helps to evaluate the validity of financial processes in the organization. The main issue is that organizations have limited data about how their financial processes run. Based on expert knowledge, normative patterns of financial process types can be developed. Normative rules can be quite complex, and difficult to understand, even if they are systematized in tables or text descriptions. The aim of the article is to present the possibilities of the Process Mining (PM) technology to discover a model of the normative financial process (by the example of the Expenditure cycle). The primary data in this kind of PM project is a list of the meta-events indicating allowed transitions between financial transaction entities (journal types, document types, account names, etc.), i.e. this meta-event-log.

The result of PM is a visualization of the normative rules – the meta-model, convenient to analyze by an expert, to reveal properties of financial processes. The Meta-model of the normative financial process (pattern) could be further used as criteria (restriction) in analyzing financial data records and detecting anomalies in financial data. The experiment results (using the Expenditure cycle as an example) reveal the capability of using meta-models (patterns of financial transactions) in financial data analysis with PM tools.

Automatic sentiment analysis is an important technique having a significant impact on many businesses and other fields. Well known fact is that sentiments are culturally dependent phenomena and are differently expressed in various cultural groups. Successful implementation of automatic sentiment identification techniques requires using sentiment corpora. Less widely spoken languages such as Lithuanian often suffer from the lack of corpora, particularly culturally specific corpora. This paper presents the results of an evaluation of the possibilities to apply machine learning techniques and the implementation of other language text corpora for sentiment analysis of texts from representatives of Lithuanian youth subcultures. The results show that quite a high accuracy (about 80–85%) could be achieved at least in some contexts.

Educational chatbots are digital tools designed to assist learners in various educational settings. These chatbots use natural language processing (NLP) and machine learning algorithms to simulate human conversation and respond to user queries in a way that facilitates learning. They can be integrated into various educational platforms such as learning management systems, educational apps, and websites to provide learners with a personalized and interactive learning experience. Our paper discusses different scenarios for educational purposes and suggests in total four scenarios for educational needs.

This paper presents a comprehensive analysis of user perspectives on an educational game designed to promote civic and social inclusion. The study employed a questionnaire-based survey with 302 respondents, aimed at gathering insights into the players’ experiences, perceptions, and attitudes towards the game. The survey explored various aspects such as game mechanics, educational content, user engagement, and the potential impact on civic and social awareness. The results of the study indicated a generally positive reception of the educational game among the respondents. The majority reported finding the game engaging and enjoyable, with a high level of immersion and interactivity. The educational content was deemed informative and relevant, contributing to the players’ understanding of civic and social issues. Furthermore, the game was observed to foster empathy and perspective-taking, enhancing the players’ ability to appreciate diverse viewpoints. Overall, this research sheds light on the user perspectives regarding an educational game for achieving societal changes. The findings highlight the game’s potential as an effective tool for promoting civic awareness, social empathy, and inclusive education. The insights gained from this study can inform the future development of similar educational games, aiding in the design of more engaging and impactful experiences that facilitate civic and social learning among diverse user populations.

Quantum Computing is envisioned as one of the scientific areas with greater transformative potential. Already there exist applications running in quantum devices for different areas, like cybersecurity, chemistry, or machine learning. One subarea being developed under quantum machine learning is quantum natural language processing. Following the promising results existing in problems like sentiment classification or next-word prediction, this paper presents two proofs of concept to demonstrate how these two tasks can be solved using quantum computing. For the first task showcased, sentiment classification, we employ the removal of caps and cups morphisms to make the string diagrams simpler and more efficient. In the case of next-word prediction, we show how to solve the task for sentences with previously unknown syntactic structures by applying a classical Random Forest machine learning algorithm that classifies the syntactic structure and enables our QNLP algorithm to infer the proper string model.

Principal Component Analysis (PCA) is a well-known dimensionality reduction technique that has been widely used in various machine learning algorithms. This includes kNN and Naive Bayes algorithms which can be time-consuming. The reduction of dimensions can have positive effects on those two algorithms by reducing the number of related types of data and decreasing the data they need to analyze. Here we present detailed findings about how the PCA algorithm affects them both in time efficiency and accuracy. All calculations regarding those values were carried out in Python programming language. The dataset used in research is the Titanic dataset, on which data cleaning and normalization were done. The data in this paper suggests that it is possible to maintain the same level of accuracy with great improvement in time efficiency. For the kNN algorithm reducing the number of dimensions by one resulted in a 31.09% increase in accuracy and for the Naive Bayes algorithm an 18.18% increase while having an imperceptible effect on accuracy.

The choice of the appropriate method in the classification task is most often a problem related to the adaptation of the input data to the classifier. However, adaptation alone does not result in high classification scores. In this paper, we present a comparison of two artificial intelligence methods for recognizing and classifying the handwriting of digits. The study was based on the popular MNIST database, and we dug up algorithms such as K-nearest neighbors and also neural networks to conduct the study. The paper presents mathematical models of selected tools and selected network architecture. Then, the results of the research carried out in order to choose a more accurate character classification technique are presented. For the purpose of verification, the accuracy metric and the analysis using the error matrix were used. Article also includes analysis of different variables to used methods, like metrics (Euclidean, Manhattan and Chebyshev) or hyperparameter k.

Heart disease has been the leading cause of death in the EU for many years. Early detection of this disease increases a patient’s chance of survival. The aim of the study is to see if machine learning algorithms can help in the early diagnosis of these illnesses. For this purpose, three classifiers: kNN, Naive Bayes and SVM were implemented and trained on a dataset containing medical data related to the possibility of cardiovascular disease. The result of the study is a comparative analysis of the classifiers that summarises the accuracy and stability of the results in determining the possibility of heart disease. The results show the highest accuracy and stability of the SVM classifier, which achieves an average of 82.47% accuracy in disease prediction, meaning that machine learning algorithms can significantly aid in the early diagnosis of patients based on their basic medical data.

In today’s world, where solutions from the last century are no longer enforced, there is a constant demand for newer, more efficient ways to analyze data. An example of such an application is the k-nearest neighbors (k-nn) mechanism. In this article, this mechanism will be proposed, improved by the possibility of finding the optimal number of neighbors and the coefficient m for the Minkowski function used in it to calculate the distance between points. This mechanism is automated, which allows you to use different parameters for the Minkowski function and determine the accuracy for a different number of neighbors in an automatic way. From these accuracies, the ranking system selects the best values for the parameter m, which defines the dimension of the space in the Minkowski function, and the best number of nearest neighbors. The number of nearest neighbors checked and the value of the m parameter can be set independently, which allows you to check various combinations of the m parameter and the number of nearest neighbors.

Machine learning is widely utilized across various scientific disciplines, with algorithms and data playing critical roles in the learning process. Proper analysis and reduction of data are crucial for achieving accurate results. In this study, our focus was on predicting the correlation between cigarette smoking and the likelihood of diabetes. We employed the Naive Bayes classifier algorithm on the Diabetes prediction dataset and conducted additional experiments using the k-NN classifier. To handle the large dataset, several adjustments were made to ensure smooth learning and satisfactory outcomes. This article presents the stages of data analysis and preparation, the classifier algorithm, and key implementation steps. Emphasis was placed on graph interpretation. The summary includes a comparison of classifiers, along with standard deviation and standard error metrics.

The rapid growth of IT systems that use artificial intelligence algorithms necessitates increasingly accurate methods. To handle data uncertainty, computer scientists can employ soft sets. One popular classification method in machine learning that utilizes the idea of proximity between data points is the k-NN algorithm. In this paper, we describe a modification to the k-NN algorithm that makes use of soft sets to take into account uncertainty in the classification process. This is achieved by introducing soft inference as a voting mechanism. The authors present a mathematical model with pseudocode for re-implementation purposes and demonstrate and discuss experimental results from conducted tests to show the effectiveness of the proposed approach.

In this paper an electronic correction of the symmetry of the states is proposed that determines the angular position of the BLDC motor shaft. The results illustrating the measurements of the asymmetry of the states of the Hall effect sensors determining the position of the motor shaft following from the adopted measuring system are presented. The improvement in efficacy by improving the symmetry of the power supply of the motor windings tested resulting from eliminating the asymmetry of signals from the sensors is shown.

Multi-Criteria Decision Making (MCDM) is a method that allows to make a decision based on many different factors. Such solutions are important from a practical point of view in situations where there are many important criteria to examine. This work considers a situation in which many patients suffer from multiple symptoms, and focus should be on those most in need. For this purpose, publicly available databases related to COVID-19 symptoms were used. The proposition is composed of processing different types of samples and a combination of their numerical values. Then, it is used in selected entropy-weighted MCDM methods for returning a patient’s ranking. The proposed solution shows that this approach has great potential due to the possibility of practical use.