Explore the vast range of titles available.

Effective website usability assessment is crucial for improving user experience, driving customer satisfaction, and ensuring business success, particularly in the competitive e-commerce sector. Traditional methods, such as expert reviews and user testing, are resource-intensive and often fail to fully capture the complex interplay between a site’s aesthetic design and its technical performance. This paper introduces an end-to-end multimodal deep learning framework that automates the usability assessment of fashion e-commerce websites. The framework fuses structured numerical indicators (e.g., load time, mobile compatibility) with high-level visual features extracted from full-page screenshots. The proposed framework employs a comprehensive set of visual backbones—including modern architectures such as ConvNeXt and Vision Transformers (ViT, Swin) alongside established CNNs—and systematically evaluates three fusion strategies: early fusion, late fusion, and a state-of-the-art cross-modal fusion strategy that enables deep, bidirectional interactions between modalities. Extensive experiments demonstrate that the cross-modal fusion approach, particularly when paired with a ConvNeXt backbone, achieves superior performance with a 0.92 accuracy and 0.89 F1-score, outperforming both unimodal and simpler fusion baselines. Model interpretability is provided through SHAP and LIME, confirming that the predictions align with established usability principles and generate actionable insights. Although validated on fashion e-commerce sites, the framework is highly adaptable to other domains—such as e-learning and e-government—via domain-specific data and light fine-tuning. It provides a robust, explainable benchmark for data-driven, multimodal website usability assessment and paves the way for more intelligent, automated user-experience optimization.

Traffic flow prediction is a pivotal element in Intelligent Transportation Systems (ITSs) that provides significant opportunities for real-world applications. Capturing complex and dynamic spatio-temporal patterns within traffic data remains a significant challenge for traffic flow prediction. Different approaches to effectively modeling complex spatio-temporal correlations within traffic data have been proposed. These approaches often rely on a single model to capture temporal dependencies, which neglects the varying influences of different time periods on traffic flow. Additionally, these models frequently utilize either static or dynamic graphs to represent spatial dependencies, which limits their ability to address complex and overlapping spatial relationships. Moreover, some approaches struggle to fully capture spatio-temporal variations, leading to the exclusion of critical information and ultimately resulting in suboptimal prediction performance. Thus, this paper introduces the Adaptive Spatio-Temporal Attention-Based Multi-Model (ASTAM), an architecture designed to capture spatio-temporal dependencies within traffic data. The ASTAM employs multi-temporal gated convolution with multi-scale temporal input segments to model complex non-linear temporal correlations. It utilizes static and dynamic parallel multi-graphs to facilitate the modeling of complex spatial dependencies. Furthermore, this model incorporates a spatio-temporal self-attention mechanism to adaptively capture the dynamic and long-term spatio-temporal variations in traffic flow. Experiments conducted on four real-world datasets reveal that the proposed architecture outperformed 13 baseline approaches, achieving average reductions of 5.0% in MAE, 13.28% in RMSE, and 6.46% in MAPE across four datasets.

Advanced metering infrastructures (AMIs) aim to enhance the efficiency, reliability, and stability of electrical systems while offering advanced functionality. However, an AMI collects copious volumes of data and information, making the entire system sensitive and vulnerable to malicious attacks that may cause substantial damage, such as a deficit in national security, a disturbance of public order, or significant economic harm. As a result, it is critical to guarantee a steady and dependable supply of information and electricity. Furthermore, storing massive quantities of data in one central entity leads to compromised data privacy. As such, it is imperative to engineer decentralized, federated learning (FL) solutions. In this context, the performance of participating clients has a significant impact on global performance. Moreover, FL models have the potential for a Single Point of Failure (SPoF). These limitations contribute to system failure and performance degradation. This work aims to develop a performance-based hierarchical federated learning (HFL) anomaly detection system for an AMI through (1) developing a deep learning model that detects attacks against this critical infrastructure; (2) developing a novel aggregation strategy, FedAvg-P, to enhance global performance; and (3) proposing a peer-to-peer architecture guarding against a SPoF. The proposed system was employed in experiments on the CIC-IDS2017 dataset. The experimental results demonstrate that the proposed system can be used to develop a reliable anomaly detection system for AMI networks.

Non-functional requirements (NFRs) are critical for ensuring software quality, yet they remain difficult to identify due to their implicit and loosely defined relationship with functional requirements (FRs). Existing research predominantly focuses on NFR classification, leaving the more practical problem of linking FRs with their corresponding NFRs largely underexplored. To bridge this gap, this research introduces Re-Distill, a framework that treats FR–NFR association as a retrieval task. It adopts a curriculum-guided, data-centric distillation strategy to improve semantic representations and capture the interdependencies between FRs and NFRs. The framework combines general semantic adaptation, domain-specific specialization, and teacher-guided hard-negative mining in a contrastive learning setting. During inference, it integrates dense and lexical retrieval with cross-encoder reranking to produce ranked NFR candidates for unseen FR queries. Experiments on an expanded FR–NFR dataset show consistent improvements throughout all training stages. The resulting model achieves a Recall@10 of 70.79%, significantly outperforming the zero-shot baseline (42.36% Recall@10). These results highlight the effectiveness of retrieval-based approaches for functional–non-functional requirement linking, providing a practical and scalable way to undertake software requirement analysis.

In today’s competitive digital landscape, application usability plays a critical role in user satisfaction and retention. Negative user reviews offer valuable insights into real-world usability issues, yet traditional analysis methods often fall short in scalability and contextual understanding. This paper proposes an intelligent framework that utilizes large language models (LLMs), including GPT-4, Gemini, and BLOOM, to automate the extraction of actionable usability recommendations from negative app reviews. By applying prompting and fine-tuning techniques, the framework transforms unstructured feedback into meaningful suggestions aligned with three core usability dimensions: correctness, completeness, and satisfaction. A manually annotated dataset of Instagram negative reviews was used to evaluate model performance. Results show that GPT-4 consistently outperformed other models, achieving BLEU scores up to 0.64, ROUGE scores up to 0.80, and METEOR scores up to 0.90—demonstrating high semantic accuracy and contextual relevance in generated recommendations. Gemini and BLOOM, while improved through fine-tuning, showed significantly lower performance. This study also introduces a practical, web-based tool that enables real-time review analysis and recommendation generation, supporting data-driven, user-centered software development. These findings illustrate the potential of LLM-based frameworks to enhance software usability analysis and accelerate feedback-driven design processes.

Traffic flow prediction (TFP) is a fundamental component of intelligent transportation systems (ITS) that supports traffic management, congestion mitigation, and route planning. Although recent advances in deep learning have demonstrated strong capability in modeling non-linear spatio-temporal correlations, most existing approaches rely on centralized training paradigms, which incur substantial communication costs, high computational overhead, and significant data privacy risks. Federated Learning (FL) has emerged as a promising alternative by enabling decentralized model training across distributed clients while reducing privacy risks and communication overhead. However, existing FL-based TFP frameworks often employ local models with limited capacity to capture complex spatio-temporal dependencies, and their reliance on the conventional FedAvg aggregation approach restricts robustness under heterogeneous traffic data distributions. To address these challenges, this study proposes the FedASTAM framework, which integrates FL with the Adaptive Spatio-Temporal Attention-based Multi-Model (ASTAM) to effectively model complex and non-linear spatio-temporal traffic correlations in a data-local FL setting. Within FedASTAM, the road network is divided into sub-regions using spectral clustering, allowing each sub-region to train a local ASTAM model tailored to localized and heterogeneous traffic patterns. At the central server, locally trained models are aggregated using seven aggregation schemes, including the classical FedAvg, to optimize global model updates while preserving data locality. Extensive experiments conducted on two real-world benchmark datasets, PeMS04 and PeMS08, demonstrate that FedASTAM achieved strong and stable predictive performance while keeping raw data localized throughout the federated training process. The results further indicate that the aggregation approaches used in the proposed FedASTAM framework generally outperform classical FedAvg under heterogeneous traffic conditions, highlighting FedASTAM as an effective approach for traffic flow prediction in complex, distributed ITS environments.

Speech acts are a speaker’s actions when performing an utterance within a conversation, such as asking, recommending, or thanking, or making a suggestion. Understanding speech acts helps interpret the intended meaning and actions behind a speaker’s or writer’s words. This article proposes a Twitter dialectal Arabic speech act classification approach based on a transformer deep learning neural network. We proposed a BERT based weighted ensemble learning approach to integrate the advantages of various BERT models in dialectal Arabic speech acts classification. We compared the proposed model against several variants of Arabic BERT models and sequence-based models. We developed a dialectal Arabic tweet act dataset by annotating a subset of a large existing Arabic sentiment analysis dataset (ASAD) based on six speech act categories. We also evaluated the models on a previously developed Arabic Tweet Act dataset (ArSAS). To overcome the class imbalance issue commonly observed in speech act problems, a transformer-based data augmentation model was implemented to generate an equal proportion of speech act categories. The results show that the best BERT model is araBERTv2-Twitter models with a macro-averaged F1 score and an accuracy of 0.73 and 0.84, respectively. The performance improved using a BERT-based ensemble method with a 0.74 and 0.85 averaged F1 score and accuracy on our dataset, respectively.

Background/Objectives: Chronic kidney disease (CKD) remains a major public health challenge due to its silent progression and late clinical detection. Recent advances in machine learning have demonstrated promising performance in CKD detection; however, most existing approaches focus primarily on binary classification or rely on longitudinal or specialized biomarkers that are not routinely available in clinical practice. While several studies attempt risk stratification, few integrate risk modeling with stage-aware hierarchical decision frameworks suitable for routine clinical workflows. This study proposes a risk-oriented, explainable, and hierarchical machine learning framework for CKD classification using real-world laboratory data from 746 patients in a Saudi population. Methods: The proposed framework is designed as a hierarchical machine learning pipeline that mirrors clinical practice by sequentially identifying CKD presence, performing disease staging only for confirmed cases, and estimating risk for individuals without overt CKD. Specifically, an XGBoost model with recursive feature elimination (RFE) was employed for binary CKD detection, followed by a multilayer perceptron (MLP) model with SelectKBest for stage classification. A unified preprocessing pipeline, clinically informed feature selection, and validated machine learning models were employed to develop the hierarchical prediction system. Results: The system achieved 97% accuracy and F1-score in binary CKD classification, and up to 85% accuracy and 86% F1-score in stage classification. In addition, an interpretable risk scoring mechanism and SHAP-based explanations enabled early identification of CKD-like laboratory patterns using routine laboratory tests. Conclusions: The proposed system provides a transparent and deployable framework that supports preventive nephrology and clinically meaningful decision-making.

(1) Background: Globally, the kidney donor shortage has made the allocation process critical for patients awaiting a kidney transplant. Adopting Machine Learning (ML) models for donor–recipient matching can potentially improve kidney allocation processes when compared with traditional points-based systems. (2) Methods: This study developed an ML-based approach for donor–recipient matching. A comprehensive evaluation was conducted using ten widely used classifiers (logistic regression, decision tree, random forest, support vector machine, gradient boosting, boost, CatBoost, LightGBM, naive Bayes, and neural networks) across three experimental scenarios to ensure a robust approach. The first scenario used the original dataset, the second used a merged version of the dataset, and the last scenario used a hierarchical architecture model. Additionally, a custom ranking algorithm was designed to identify the most suitable recipients. Finally, the ML-based donor–recipient matching model was integrated into a web-based platform called Nephron. (3) Results: The gradient boost model was the top performer, achieving a remarkable and consistent accuracy rate of 98% across the three experimental scenarios. Furthermore, the custom ranking algorithm outperformed the conventional cosine and Jaccard similarity methods in identifying the most suitable recipients. Importantly, the platform not only facilitated efficient patient selection and prioritisation for kidney allocation but can be flexibly adapted for other solid organ allocation systems built on similar criteria. (4) Conclusions: This study proposes an ML-based approach to optimize donor-recipient matching within the kidney allocation process. Successful implementation of this methodology demonstrates significant potential to enhance both efficiency and fairness in kidney transplantation.

App reviews are crucial in influencing user decisions and providing essential feedback for developers to improve their products. Automating the analysis of these reviews is vital for efficient review management. While traditional machine learning (ML) models rely on basic word-based feature extraction, deep learning (DL) methods, enhanced with advanced word embeddings, have shown superior performance. This research introduces a novel aspect-based sentiment analysis (ABSA) framework to classify app reviews based on key non-functional requirements, focusing on usability factors: effectiveness, efficiency, and satisfaction. We propose a hybrid DL model, combining BERT (Bidirectional Encoder Representations from Transformers) with BiLSTM (Bidirectional Long Short-Term Memory) and CNN (Convolutional Neural Networks) layers, to enhance classification accuracy. Comparative analysis against state-of-the-art models demonstrates that our BERT-BiLSTM-CNN model achieves exceptional performance, with precision, recall, F1-score, and accuracy of 96%, 87%, 91%, and 94%, respectively. The significant contributions of this work include a refined ABSA-based relabeling framework, the development of a high-performance classifier, and the comprehensive relabeling of the Instagram App Reviews dataset. These advancements provide valuable insights for software developers to enhance usability and drive user-centric application development.