While sentiment analysis is critical for monitoring the market adoption of novel foods, its practical application is often hindered by severe class imbalance and prohibitive costs of manual annotation. This study developed a data-centric machine learning (ML) framework for sentiment analysis using 4,616 online product reviews. To address severe class imbalance and high manual annotation costs, we compared self-labeling and clustering-based labeling schemes. Hybrid resampling strategies were applied across five ML algorithms (logistic regression, support vector machine, random forest, XGBoost, and CatBoost) and two fine-tuned transformer models (KcELECTRA, KoELECTRA) for binary (positive/negative) and ternary (positive/neutral/negative) classification. While KcELECTRA performed exceptionally well under ideal self-labeling conditions (Macro F1 0.83), the performance gap between deep learning and machine learning narrowed significantly in practical, automated scenarios. Under the clustering scheme, resampled ML models (CatBoost) achieved a Macro F1 of 0.69, effectively matching or even surpassing the fine-tuned large language model (Macro F1 0.65), which proved more sensitive to label noise. These results demonstrate that a well-calibrated ML pipeline, enhanced by hybrid resampling, offers highly competitive predictive power with only a fraction of the computational cost and manual labor. Consequently, this study suggests that for domain-specific sentiment analysis, where expert labels are scarce, the robustness–efficiency trade-off of the proposed ML framework provides a more viable and sustainable alternative to heavyweight transformer architectures. This data-centric approach democratizes large-scale, real-time consumer-perception monitoring for resource-constrained emerging food sectors.
The insulin signaling pathway, involving protein kinase B (PKB) and mitogen-activated protein kinase (MAPK), mediates the biological response to insulin and several growth factors and cytokines. To investigate the correlation between glucose transporter (Glut) biosynthesis and the insulin signaling pathway activated by novel compounds of Liriope platyphylla (LP9M80-H), alterations in Glut and key protein expression in the insulin signaling pathway were analyzed in the liver and brain of ICR mice treated with LP9M80-H. An in vitro assay showed that the highest level of insulin concentration was observed in the LP9M80-H-treated group, followed by the LP-H, LP-M, LP-E, and LP9M80-C-treated groups. Therefore, LP9M80-H was selected for use in studying the detailed mechanism of the insulin signaling pathway in animal systems. In an in vivo experiment, LP9M80-H induced a significant increase in glucose levels and a decrease of insulin concentration in the blood of mice, while their body weight remained constant over 5 days. The expression level of Glut-3 was down-regulated in the liver, or maintained at the same level in the brain of LP9MH80-H-treated mice. These changes corresponded to the phosphorylation of the p38 protein rather than to ERK and JNK in the MAPK signaling pathway. In addition, the expression level of Glut-1 increased significantly after LP9MH80-H treatment of both insulin target tissues in mice. Western blot analysis showed that Akt in the PI3-K pathway mainly participated in Glut-1 biosynthesis. Thus, these results suggest the possibility that the LP9M80-H-induced regulation of Glut-1 and Glut-3 biosynthesis may be mediated by the Akt and p38 MAPK signaling of the insulin signaling pathway in the liver and brain of mice.
This study investigates the unsteady magnetohydrodynamic flow and heat-mass transfer of a Casson nanofluid through a flat surface under a variable pressure gradient and velocity-slip boundary condition. The non-Newtonian rheological behavior of the Casson fluid is considered to capture yield-stress effects in practical transport systems. The nanofluid is modeled using the Buongiorno framework, which accounts for nanoparticle transport due to Brownian motion and thermophoresis. The mathematical formulation consists of coupled momentum, energy, and concentration equations, incorporating the effects of viscous dissipation, Joule heating, and a homogeneous chemical reaction. The nonlinear governing equations are solved numerically using a finite-difference method combined with the successive over-relaxation (SOR) technique. The results indicate that increasing the Casson parameter enhances the velocity distribution, whereas a higher slip parameter reduces the near-wall velocity. Moreover, Brownian motion increases the fluid temperature, while thermophoresis significantly alters the concentration profile by driving nanoparticles away from the heated surface. These findings provide useful insight into the combined influence of rheology, magnetic forces, slip effects, and pressure variation on transport behavior, which can support the design of channel systems in thermal and industrial engineering applications. From a sustainability perspective, the enhanced thermal performance and controllable transport characteristics of nanofluids can contribute to energy-efficient thermal management systems, reduced energy consumption, and improved design of environmentally sustainable industrial and engineering processes.
This study proposes a multi-objective optimization model for designing low-carbon transportation routes for perishable agricultural products, integrating environmental sustainability, economic efficiency, and customer satisfaction. To address the limitations of conventional ant colony algorithms-such as local optima entrapment and slow convergence-we introduce an improved ant colony algorithm (IACO) enhanced with a dynamic pheromone update mechanism and a risk-aware 2-opt local search strategy. The model simultaneously minimizes total cost, carbon emissions, and product loss while maximizing customer satisfaction, incorporating realistic constraints such as vehicle capacity, time windows, and cold-chain temperature-humidity requirements. A simulation-based case study of a Chinese agricultural wholesale center serving 20 supermarkets demonstrates that the proposed IACO reduces carbon emissions by 19% (to 73.08 kg) and improves customer satisfaction to 64%, outperforming particle swarm optimization (31%) and simulated annealing (32%). Moreover, 80% of routes are confined within 100 km, achieving a synergistic balance of short routes, low spoilage, and low emissions. The research provides a computationally efficient and scenario-adaptive framework for green coldchain logistics, contributing to sustainable agri-food supply chain engineering.
In the global context, cross-cultural knowledge integration is the key issue of knowledge management. This study takes the integration of North and Western learning into Russian films as a case to explore the integration mechanism of heterogeneous cultural knowledge and its influence on narrative paradigm. It is found that the Nordic elements such as simple representation and naturalism have injected a new cognitive dimension into Russian traditional narrative. This paper constructs a cross-cultural narrative knowledge integration model, focusing on the knowledge coding of introspective emotional expression and deep psychological depiction, and systematically explains how it reconstructs narrative logic and audience cognitive path. This research model reveals the co-evolution mechanism of representation, theme, and emotional resonance in cross-cultural knowledge transfer. This study expands the modeling, migration, and application of knowledge management to unstructured cultural knowledge, and provides reference for cross-cultural content generation and narrative system design.
