The massive growth of electronic data has created a demand for efficient tools to manage information and support fast decision-making. Automatic text summarization (ATS) addresses this by condensing large texts into concise, relevant summaries rapidly. ATS methods are categorized as extractive, abstractive, or hybrid. Extractive techniques select key sentences from input documents, whereas abstractive techniques generate new sentences to capture meaning. Hybrid methods combine both strategies. However, despite numerous suggested techniques, machine-generated summaries often fail to match the accuracy and coherence of human-written summaries. This study reviewed existing ATS techniques and highlighted their limitations, particularly high computational costs and low training efficiency. To address these problems, this study proposed an improved multilayer extreme learning machine autoencoder (MLELM-AE) and an ensemble learning framework that integrates four machine learning models: Sentence-bidirectional encoder representations from transformers, autoencoder, variational autoencoder, and the improved MLELM-AE. The proposed framework generates summaries through cosine similarity evaluation, followed by voting-based fusion, re-ranking, and optimal sentence selection. Experimental results showed that the proposed improved MLELM-AE model achieved strong performance, with an execution time of 50,015 ms and a recall-oriented understudy for gisting evaluation 1 score of 0.865145. These findings demonstrate that the proposed ensemble framework delivers more accurate and efficient summaries, offering a promising advancement in ATS.

The importance of military catering in military organizations cannot be overlooked, as it not only impacts the health and physical fitness of service members but also directly affects combat readiness and morale. This study focuses on a northern air force base, using the Parasuraman-Zeithaml-Berry service quality (SERVQUAL) model’s Gap 1 and Gap 5 as its framework. The aim is to investigate the perception gaps in catering service quality between food service providers and customer. An importance-performance analysis matrix is employed to further analyze the findings. The analysis reveals that, regarding “catering service quality,” food service providers who are actively serving without formal food service certification, and those with high school or college education, tend to place more emphasis on tangibility, reliability, empathy, and responsiveness. For service quality expectations, customers who possess a college education and have obtained a food service certification show higher expectations in tangibility and reliability dimensions. Younger customer, aged 18-25, who are uncertified and less experienced, report greater satisfaction with the catering service’s reliability, responsiveness, and assurance dimensions after their experience with the base’s services. Regarding the perception difference in Gap 1 of the SERVQUAL model, the study suggests that services should prioritize user experience and ensure transparency by publicizing findings from meal review meetings. Feedback can be gathered through a satisfaction mailbox to address and efficiently amend any service deficiencies. For Gap 5 in terms of experience, customers show particular concern for food safety measures and overall service quality, indicating that these areas should be maintained or enhanced. Regular training is recommended to improve the knowledge and effectiveness of food service providers in these critical aspects.

Graphics processing units (GPUs) have emerged as powerful platforms for parallel computing, enabling personal computers to solve complex optimization tasks effectively. Although swarm intelligence algorithms naturally lend themselves to parallelization, a GPU-based implementation of the simplified swarm optimization (SSO) algorithm has not been reported in the literature. This paper introduces a compute CUDA-SSO algorithm on the CUDA platform, with a time complexity analysis of O (Ngen × Nsol × Nvar), where Ngen is the number of iterations, Nsol is the population size (i.e., number of fitness function evaluations), and Nvar represents the required pairwise comparisons. By eliminating resource preemption of personal best and global best updates, CUDA-SSO significantly reduces the overall complexity and prevents concurrency conflicts. Numerical experiments demonstrate that the proposed approach achieves an order-of-magnitude improvement in run time with superior solution precision relative to central processing unit-based SSO, making it a compelling methodology for large-scale, data-parallel optimization tasks.

The development of microarray technology has facilitated expression profiling analysis for various medical and agricultural research areas. Despite the increasing range of applications, precision in isolating microarray images remains a challenge due to noise and variances in spot morphology. This research proposes a hybrid and adaptive clustering solution that offers significant improvement in terms of accuracy, segmentation, noise reduction, processing time, and overall efficiency. The study used an adaptive K-means clustering approach enhanced with genetic algorithms and bi-dimensional empirical mode decomposition. This hybrid framework achieved an average segmentation accuracy of approximately 95%, compared to 85% with conventional K-means, showing its superiority. In addition, the enhanced method achieved unparalleled noise reduction by 80% and improved signal-to-noise ratio by 200%, while maintaining efficiency with an average image processing time of 1.2 s. These results uniquely address complex challenges in microarray image analysis, unlocking new solutions critical for gene profiling in medicine and agriculture, and driving transformative advancements in the sectors.

Wireless sensor networks (WSNs) face critical challenges in fault detection that can compromise their quality of service in dynamic environments. This study introduces an integrated framework that enhances fault detection by combining advanced noise filtering, optimized feature selection, and a robust deep learning (DL) model. The framework employs a dynamic noise filtering technique with adaptive thresholding to effectively remove noise while preserving essential data integrity. Complementing this, the rank-based whale optimization algorithm refines feature selection, boosts model performance, and reduces computational demands. At its core, the hierarchical attention-based DL model utilizes temporal convolutional layers, long short-term memory units, and hierarchical attention mechanisms to capture both short-term and long-term dependencies in the data. Experimental evaluations on the WSN dataset demonstrate outstanding performance, with a precision of 0.98, a recall of 0.99, an F1-score of 0.98, and an area under the curve of 0.99 for all fault classes. Comparative analysis reveals that this framework outperforms existing approaches in terms of accuracy, sensitivity, specificity, and computational efficiency. Overall, the proposed solution improves fault detection and enhances network reliability, minimizes false alarms, and extends the operational lifespan of WSNs, offering a scalable approach for mission-critical applications in healthcare, environmental monitoring, and industrial automation.