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: Some of the world's most valuable platform businesses rely on products and services provided by small and medium-sized enterprises (SMEs).Though, the modern digital platform economy is increasingly shaped by uncertainties and power asymmetries benefitting dominant platform owners and threatening smaller players participating as complementors in those ecosystems.Negative consequences include lock-in effects and platform dependency, exploitative participation terms and eroded entrepreneurial autonomy on the SMEs' side, which altogether harm the digital platforms' longterm viability, too.Addressing these issues, this paper investigates design principles for digital platforms taking SME complementors' needs into account.Objective: This study investigates design principles for digital platforms that enhance suitability for SMEs as complementors, focusing on stakeholder-centric platform design approaches that better accommodate SME-specific needs and requirements.In this, this study aims to address current negative developments concerning imbalanced power dynamics and uncertainties emerging from platform owner-SME-partnerships in dominant digital platform ecosystems.Methods: A qualitative reflective meta-analysis approach was employed, combining explorative expert interviews with SME specialists and a thematic literature review of SME platform design research.The methodology synthesized findings to identify meta-requirements and derive design principles through interpretive analysis.Results: Eleven meta-requirements were identified and synthesized into four design principles: the principle of SME empowerment, the principle of open boundaries, the principle of transparent and fair participation terms, and the principle of reflection of individuality.In combination, these principles aim to inform future digital platform design that takes SME complementor needs into account. Conclusion:The conceptual proposal of four design principles guides researchers and practitioners in creating SME-suitable digital platforms with stakeholder-centric design approaches.The principles enable digital platform models that accommodate diverse SME requirements, enhance participation experiences, and foster collaborative ecosystems tailored to SME characteristics and their operational contexts.

Background: Rapid technological advancements have revolutionized research into unmanned aerial vehicles (UAVs), commonly known as drones, particularly in delivery applications.However, despite numerous related publications, there remains a lack of systematic reviews that synthesize challenges, trends and recent advances in drone delivery.To address this gap, the present study conducts a bibliometric analysis to examine evolutionary trends and emerging applications of UAVs between 2015 and 2024.Objective: This study aims to identify established and emerging trends in drone delivery research by analysing articles, journals, authors, institutions, countries and thematic areas.Methods: Previous studies are selected using a systematic approach, followed by bibliometric analysis with tools including VOSviewer, Bibliometrix and ScientoPy, which emphasizes key authors, top journals and countries, collaboration patterns and recurring author keywords.Results: The bibliometric analysis of 1,438 articles from 583 sources authored by 4,333 scholars (2015-2024) reveals a strong interdisciplinary focus in drone delivery research.Military applications largely drove early studies, but recent breakthroughs highlight the integration of artificial intelligence (AI) for autonomous navigation and energy optimization.Emerging themes include the development of drone swarms for scalable applications such as disaster response and agricultural mapping.Geographically, China, the United States and Australia dominate contributions, with extensive international collaborations fostering global innovation.Across journals and authors, the literature reflects a steady evolution from conceptual and technical foundations to applied studies addressing logistics, smart cities and environmental monitoring.Overall, the results suggest that drone delivery research is transitioning from exploratory phases towards AI-enabled autonomy deployment.Conclusion: Drone delivery research has evolved from military origins into a global, interdisciplinary field driven by AI.China, the USA and Australia are the leading contributors.Its future hinges on balancing technological innovation-such as autonomous navigation and swarm applications-with ethical, regulatory and societal considerations for sustainable integration.

Background: Benford's law is a statistical phenomenon that predicts the probability of a particular digit at a particular position in a number. This law has been successfully applied in a number of areas, such as accounting. In the area of scientometrics, research has been devoted mostly to journal data.Objective: This paper investigates the conformity of Benford's law with the citation counts of records retrieved from the Web of Science database. We evaluate the conformity levels with Benford's law in the complete dataset. We determine the effect of document type (article, proceedings paper and review), year of publication (2014-2018) and Web of Science categories (254 categories) on the level of conformity of the citation counts with Benford's law.Methods: The dataset of this research contains over 8.47 million records. All available records from the Web of Science were downloaded, so this set is the entire population of data available at the time of download. The distributions of the first significant digits in the citation counts of these records are compared with Benford's law. Mean absolute deviation (MAD) recommended by Nigrini (2012) and sum of squared deviations (SSD) recommended by Kossovsky (2015) are used to categorize the similarity of the citation counts to Benford's law.Results: The entire dataset of this study shows marginal conformity according to both MAD and SSD intervals (with a MAD value of 0.1257 and an SSD value of 29.9; a lower value indicates a better agreement). The review document type shows a high level of conformity, while proceedings paper shows a lower level. We found significant differences in conformity between Web of Science categories.Conclusion: This study mapped the level of conformity of the citation counts with Benford's law in data from the Web of Science database. Further directions for possible research are suggested.

Background: Large language models (LLMs) excel at various tasks but often encounter difficulties when extended reasoning requires maintaining a consistent internal state. Identifying the threshold at which these systems fail under increasing task complexity is essential for reliable deployment. Objective: The primary objective was to examine whether four LLMs (GPT 3.5, GPT 4, GPT 4o-mini and GPT 4o) could preserve a hidden number and its arithmetic transformation across multiple yes/no queries and to determine whether a specific point of reasoning breakdown exists. Methods: A modified "Think a Number" game was employed, with complexity defined by the number of sequential yes/no queries (ranging from 1 to 9 or 11). Seven prompting strategies, including chain-of-thought variants, counterfactual prompts and few-shot examples, were evaluated. Each outcome was considered correct if the revealed number and transformation of the model remained consistent with prior answers. Results: Analysis of tens of thousands of trials showed no distinct performance cliff up to 9-11 queries, indicating that modern LLMs are more capable of consecutive reasoning than previously assumed. Counterfactual and certain chain-of-thought prompts outperformed simpler baselines. GPT 4o and GPT 4o-mini attained higher overall correctness, whereas GPT 3.5 and GPT 4 more often displayed contradictory or premature disclosures. Conclusion: In a controlled, scalable reasoning scenario, these LLMs demonstrated notable resilience to multi-step prompts. Both prompt design and model selection significantly influenced performance. Further research involving more intricate tasks and higher query counts is recommended to delineate the upper boundaries of LLM internal consistency.

Background:The study investigates the emerging concept of Last Mile Logistics (LML), focusing on optimization routing for e-commerce deliveries.This research addresses the growing demand for efficient delivery systems driven by the rise of online shopping and technological advancements.Objective: This article aims to analyse existing literature, identify trends and highlight gaps in LML research.Methods: A systematic approach was used for the selection of studies and then bibliometric analysis was used to map and evaluate relevant studies.The Scopus citation database was used to gather peerreviewed articles from 1999 to 2023.R-Studio and VOS Viewer were employed for data analysis and visualization.The initial search yielded 3,898 articles, which were filtered down to 1,606 articles.Key findings include the identification of the most productive journals, top-cited articles and influential keywords in LML research. Results:The study significantly emphasizes variables such as distance, costs, energy consumption and delivery time in logistics optimization.E-commerce emerged as a predominant topic, with substantial discussions on technology and transportation within LML.The research underscores the critical role of LML in the success of e-commerce. Conclusion:The findings highlight the need for ongoing innovation and optimization in delivery logistics to enhance efficiency and meet consumer demands.Future research should address the gaps identified, particularly in the context of emerging technologies such as autonomous vehicles and drones, to further improve last mile delivery systems.

Blackout events in smart grids can have significant impacts on individuals, communities and businesses, as they can disrupt the power supply and cause damage to the grid. In this paper, a new proactive approach to an early warning system for predicting blackout events in smart grids is presented. The system is based on deep learning models: convolutional neural networks (CNN) and deep self-organizing maps (DSOM), and is designed to analyse data from various sources, such as power demand, generation, transmission, distribution and weather forecasts. The system performance is evaluated using a dataset of time windows and labels, where the labels indicate whether a blackout event occurred within a given time window. It is found that the system is able to achieve an accuracy of 98.71% and a precision of 98.65% in predicting blackout events. The results suggest that the early warning system presented in this paper is a promising tool for improving the resilience and reliability of electrical grids and for mitigating the impacts of blackout events on communities and businesses.

Artificial intelligence methods, especially deep learning, have seen increasing application in analysing personality and occupational data to identify individuals with psychological and neurological disorders. Currently, there is a great need for effectively processing mental healthcare with the integration of artificial intelligence such as machine learning and deep learning. The paper addresses the pressing need for accurate and efficient methods for diagnosing psychiatric disorders, which are often complex and multifaceted. By exploiting the power of convolutional neural networks (CNN), we propose a novel CNN-based natural language processing method without removing stop words for predicting psychiatric diagnoses capable of accurately classifying individuals based on their psychological data. Our proposal is based on keeping a richer linguistic and semantic context to accurately predict psychiatric diagnosis. The experiment involves two datasets: one gathered from a private clinic and the other from Kaggle, called the Human Stress Dataset. The outcomes from the first dataset demonstrate a remarkable accuracy rate of 98.51% when employing CNN, showcasing their superior performance compared to the standard machine learning techniques such as logistic regression, k-nearest neighbours and support vector machines. With the second dataset, our model achieved an impressive area under the receiver operating characteristic curve (AUROC) of 0.87. This result surpasses those achieved by existing state-of-the-art methods, further highlighting the efficacy of our CNN-based approach in discerning subtle nuances within the data and making accurate predictions. Moreover, we have compared our model with three other programs on the same dataset and the accuracy reached 78.52%. The results are promising to aid parents or clinicians in early and rapidly predicting the ill individual.

Person identification through chest X-ray radiographs stands as a vanguard in both healthcare and biometrical security domains.In contrast to traditional biometric modalities, such as facial recognition, fingerprints and iris scans, the research orientation towards chest X-ray recognition has been spurred by its remarkable recognition rates.Capturing the intricate anatomical nuances of an individual's rib cage, lungs and heart, chest X-ray images emerge as a focal point for identification, even in scenarios where the human body is entirely damaged.Concerning the field of deep learning, a paradigm is exemplified in contemporary generations, with promising outcomes in classification and image similarity challenges.However, the training of convolutional neural networks (CNNs) requires copious labelled data and is time-consuming.In this study, we delve into the rich repository of the NIH ChestX-ray14 dataset, comprising 112,120 frontal-view chest radiographs from 30,805 unique patients.Our methodology is nuanced, employing the potency of Siamese neural networks and the triplet loss in conjunction with refined CNN models for feature extraction.The Siamese networks facilitate robust image similarity comparison, while the triplet loss optimizes the embedding space, mitigating intraclass variations and amplifying inter-class distances.A meticulous examination of our experimental results reveals profound insights into our model performance.Noteworthy is the remarkable accuracy achieved by the VGG-19 model, standing at an impressive 97%.This achievement is underpinned by a well-balanced precision of 95.3% and an outstanding recall of 98.4%.Surpassing other CNN models utilized in our research and outshining existing state-of-the-art models, our approach establishes itself as a vanguard in the pursuit of person identification through chest X-ray images.

The pursuit of fairness in machine learning models has become increasingly crucial across various applications, including bank loan approval and face detection. Despite the widespread use of artificial intelligence algorithms, concerns persist regarding biases and discrimination within these models. This study introduces a novel approach, termed "The Fairness Stitch" (TFS), aimed at enhancing fairness in deep learning models by combining model stitching and training jointly, while incorporating fairness constraints. We evaluate the effectiveness of TFS through a comprehensive assessment using two established datasets, CelebA and UTKFace. The evaluation involves a systematic comparison with the existing baseline method, fair deep feature reweighting (FDR). Our analysis demonstrates that TFS achieves a better balance between fairness and performance compared to the baseline method (FDR). Specifically, our method shows significant improvements in mitigating biases while maintaining performance levels. These results underscore the promising potential of TFS in addressing bias-related challenges and promoting equitable outcomes in machine learning models. This research challenges conventional wisdom regarding the efficacy of the last layer in deep learning models for debiasing purposes. The findings suggest that integrating fairness constraints into our proposed framework (TFS) can lead to more effective mitigation of biases and contribute to fairer AI systems.