This study investigates the impact of virtual influencers (VIs) on impulsive buying behaviour in e-commerce live streaming (ELS), focusing on the roles of expectation confirmation, parasocial relationships, and influencer-product fit. Drawing from expectation-confirmation theory and parasocial relationship theory, data from 820 valid questionnaires were analysed using partial least squares structural equation modelling (PLS-SEM) and multi-group analysis to compare the effects of human-like and anime-like VIs. The findings reveal that expectation confirmation significantly enhances satisfaction, which drives continuous watching intention and impulsive buying behaviour. Parasocial relationships positively influence satisfaction and continuous watching intention but do not moderate the link between confirmation and satisfaction, while influencer-product fit significantly affects satisfaction and continuous watching intention, partially moderating the confirmation-satisfaction relationship. Notably, anime-like VIs demonstrate stronger effects of parasocial relationships and influencer-product fit on satisfaction and continuous watching intention, respectively, compared to human-like VIs. These findings offer novel insights into the role of VIs in ELS and provide practical guidance for marketers to align VI characteristics with product promotion strategies to enhance engagement and spur impulsive purchases.
This systematic review evaluates the performance of electrochemical biosensors in detecting foodborne pathogens, synthesizing findings from 77 studies (1998-2025). While advancements in nanomaterial-based electrodes, bioreceptor selection, and signal amplification techniques have significantly improved sensitivity and specificity, their real-world applicability remains limited. A key limitation identified is the lack of real-world sample validation, with only 1 out of 77 studies conducting direct testing on naturally contaminated food matrices. The over-reliance on spiked samples and pre-enriched bacterial cultures raises concerns about biosensor reliability in uncontrolled food environments. To bridge this gap, future research should emphasize standardisation of validation protocols, real-world testing, and integration with digital technologies such as internet of things (IoT) and artificial intelligence (AI) for real-time food safety monitoring. Establishing regulatory alignment with International Organisation for Standardisation (ISO), Food and Agriculture Organisation (FAO), and Food and Drug Administration (FDA) standards will be crucial in transitioning biosensors from laboratory development to practical deployment in food safety monitoring systems.
Aligned with the United Nations Sustainable Development Goal 4 (Quality Education), this paper addresses the ethical use of Artificial Intelligence in academic writing. While AI tools offer unprecedented opportunities for personalized learning and efficiency, their unregulated use raises significant ethical challenges. Through a systematic conceptual review, this study synthesizes existing ethical frameworks and identifies gaps in practical guidance. It proposes a novel, student-centered framework grounded on four pillars: (1) Validation and Source Verification, ensuring AI-generated content accuracy; (2) Transparency and Acknowledgment, promoting honesty by disclosing AI assistance; (3) Context-Dependent Similarity Index, adopting flexible thresholds for AI content usage depending on the academic context; and (4) Direct Source Linking, encouraging proper citation practices with accessible references. The study advocates for embedding this framework into educational policies and AI literacy curricula to empower students to harness AI responsibly, thereby safeguarding educational quality and fostering a culture of transparency and ethical scholarship.
Thoracic diseases often involve multiple coexisting pathologies with significant similarities, making multi-label classification of chest X-rays a complex and challenging task. Additionally, the varying sizes of lesions, ranging from small nodules to extensive infiltration, demand robust multi-scale feature extraction while ensuring model efficiency and classification accuracy. This study proposes MSA-Net, a novel architecture built upon DenseNet-121 to enhance multi-label classification performance. The model introduces a dual-path convolutional design to achieve multi-scale feature extraction, effectively capturing both localized and global pathological patterns. Additionally, a scale-aware attention module was developed, combining channel and spatial attention to dynamically refine feature representations across different scales. A hybrid pooling strategy, which integrates global average pooling and global max pooling, is employed in the classification layer to further improve feature aggregation. The proposed model was evaluated on the ChestX-ray14 dataset with a patient-level data split. It achieved a competitive average AUC of 0.823, demonstrating its effectiveness in multi-label thoracic disease classification. This study highlights the potential of multi-scale and attention-based designs to advance automated thoracic disease diagnosis while achieving a trade-off between performance and model complexity.
Efficient path planning is essential for mobile robots to navigate from a start to a goal position while avoiding obstacles. Particle Swarm Optimization (PSO) is widely used due to its strong search capabilities, but its standard form suffers from slow convergence and local optima trapping, limiting its performance in complex environments. To address these challenges, this paper proposes an Alpha-Beta Guided Particle Swarm Optimization (ABGPSO) algorithm, incorporating two coefficients, alpha and beta, which utilize a time-varying sigmoid function to dynamically adjust particle movements. This enhancement improves PSO's navigation efficiency, ensuring smoother, collision-free paths while optimizing both travel time and distance. Experiments were carried out in four different layouts related to path-planning environments, and comparisons were made with various existing path-planning algorithms. Through extensive simulations across various static environment maps, we demonstrate that the ABGPSO algorithm outperforms existing state-of-the-art optimization techniques, including Genetic Algorithms (GA), Grey Wolf Optimization (GWO), and modern optimizers like the Sine Cosine Algorithm (SCA), Harris Hawks Optimization (HHO) and Reptile search algorithm (RSA). The results reveal that our proposed method reduces the mobile robot's travel time by up to 69%, 67%, 72%, and 79% compared to these algorithms, while consistently achieving optimal path lengths. This research contributes to the advancement of mobile robot navigation by providing a novel PSO modification that effectively balances the critical factors of distance, time, and safety in path planning. The results showed that the proposed ABGPSO algorithm reduces the time mobile robots take from start to goal.
