The technique is designed to optimize the energy consumption for sensor processing layer in Internet Of Things (IoT). Sleep time calculation algorithm is built on the gateway to predict the sleep time of sensor nodes according to information quality, information history, and remaining battery capacity. The power reduction techniques are designed to turn off the functions of node sensors in sleep mode. Sensor Node consumes 2.84% of battery power compared to 12.52% of reading data in a given cycle in 2 hours. Transmission results are 100% successful rate with distances less than 500m. The sensor node consumes 4.47mAh to 28.8mA, corresponding to one AA battery that can sustain the node from 15.6 hours to 11 days. The study also calculates the solar cell solution provided for sensor nodes.

The study proposes a hybrid data driven clock gating and data gating technique which is applied to ALU in RISC-V. By doing so, the proposed low power technique can improve the power saving efficiency. The proposed low power technique is compared with various low power techniques such as latch-free based clock gating, latch-based clock gating, single data driven clock gating, and single data gating. The results show that the proposed low power ALU saves 46.67% power consumption compared to original ALU. The proposed ALU also shows better saving power than the latch-free based clock gating, latch-based clock gating, sdata driven clock gating, and data gating from 10.84% to 22.23%. The comparison is also implemented on CPU which consists of memory, ALU and control unit. The proposed low power CPU saves 12.11% at least compared to the original CPU. However, the proposed low power CPU is reduced to 15.1% maximum frequency operation compared to the original CPU. The area overhead of the proposed ALU also increased to 33 LUTS (8.2%) and 2 registers (1.6%) compared to the original ALU.

Online shopping is extremely popular and gradually replacing traditional shopping. Online markets are increasingly appearing and becoming very popular after the COVID-19 pandemic phenomenon worldwide. This article focuses on changes in customers' online shopping behavior amid the COVID-19 pandemic. We conduct systematic scientific analysis and visualization of 407 Scopus-indexed articles from 2019 to 2024. The five-year timeframe strikes a balance between capturing meaningful trends and practicality in data collection and analysis considering the context of the COVID pandemic. N-gram analysis identifies the terms to show the development trends of online shopping consumer behavior over the years. Based on prominent keywords, top authors, institutions, journals in the field, top author keywords, co-author networks, co-keywords, and top articles are also analyzed. The relations among top keywords, top authors and top journals are conducted by the Sankey diagram analysis. The research conducts document-term-matrix (DTM) analysis to identify seven key clusters, research hotspots, and emerging trends in the bibliometric network. This study can guide further research when artificial intelligence technology integrated into internet and mobile platforms is growing strongly. The obtained research results can serve policymakers, researchers, and practitioners to apply high technology to increase consumer experience in online shopping behavior.

Type of the article: Research ArticleAbstract Online shopping behavior has been profoundly reshaped by the technological acceleration of the Fourth Industrial Revolution (IR 4.0), making it essential to understand how artificial intelligence and digital integration influence consumer decision-making. The study aims to map and quantify emerging trends in online consumer behavior research from 2019 to 2025. A bibliometric analysis was conducted on 485 Scopus-indexed articles using Python packages, incorporating keyword co-occurrence, thematic evolution, Kleinberg’s burst detection, and N-gram with Document-Term Matrix (DTM) analyses. The results reveal five dominant clusters: Consumer Behavior & Smart Retail, Intelligent Commerce, Data-Driven Behavior, Smart Experience, and Neuro-Behavioral Analytics that collectively define the field’s intellectual structure. Publication output grew dynamically, rising 19.6% in 2020, 10.9% in 2021, and peaking at 37.7% in 2022, before regaining momentum with a 29.6% increase in 2024, with an average of 19.8 citations per document. The mean burst strength (94.6%) indicates accelerated keyword activity, with the strongest surges in Technology & Data (+350%) and Shopping & Retail (+83%), followed by moderate bursts in Trust & Risk and Marketing & Communication (+67%). Thematic evolution shows 87% growth in Intelligent Commerce & Neuro-Behavioral Analytics in 2024, highlighting the growing integration of technology and consumer psychology. Keyword analysis shows a consistent rise from 2019 and peaking in 2024 with an 86.7% growth in Consumer Psychology, signaling a shift from pandemic-focused research to AI-driven and consumer-centered studies. The findings highlight the transformative impact of IR 4.0 technologies on digital consumer behavior and online commerce research.AcknowledgmentsThe deepest appreciation is expressed to University of Economics and Law, Viet Nam National University Ho Chi Minh City and Ho Chi Minh City University of Technology and Education, Hochiminh City, Vietnam.

While artificial intelligence has significantly advanced sign language recognition (SLR), high-performing models often entail substantial computational complexity. This creates a significant barrier to deploying real-time applications on resource-constrained edge hardware. While knowledge distillation (KD) offers a potential solution, existing multi-teacher strategies are predominantly tailored for static images and remain unoptimized for skeleton-based graph convolutional networks (GCNs), which are inherently sensitive to pose estimation noise and require modeling complex temporal-semantic dependencies. To address this, we propose Mask-Guided Adaptive Gating Knowledge Distillation (MAG-KD), a framework specifically tailored for skeleton-based GCNs. Our approach adopts a decoupled distillation strategy, integrating a Mask-Guided Filtering Mechanism based on correctness-aware principles to strictly filter noisy supervision, and an Adaptive Gating Network leveraging context-aware features to dynamically synthesize teacher knowledge based on input complexity. Extensive evaluations across multiple random seeds demonstrate the superior stability of our approach. On the AUTSL dataset, our lightweight MAG-KD student (Student-M, 0.48M parameters) achieves a highly robust Top-1 mean accuracy of 94.25% ± 0.19. Furthermore, our method proves strong cross-dataset generalization by achieving 66.33% ± 0.54 on the WLASL-100 subset, significantly outperforming state-of-the-art KD methods. Comprehensive hardware profiling on an NVIDIA Jetson Nano confirms optimal edge feasibility: the deployed model achieves a real-time inference speed of 33.13 frames per second (FPS) while maintaining a highly efficient power profile of merely 6.23W. Extended analysis indicates that the proposed algorithm operates optimally within the standard edge model regime, whereas static ensemble strategies remain more effective for ultra-compressed architectures with fewer than 100k parameters.