The selection of optimal text embedding models remains a critical challenge in semantic textual similarity (STS) tasks, particularly when performance varies substantially across datasets. In this study, the comparative effectiveness of multiple state-of-the-art embedding models was systematically evaluated using a benchmarking framework based on established machine learning techniques. A range of embedding architectures was examined across diverse STS datasets, with similarity computations performed using Euclidean distance, cosine similarity, and Manhattan distance metrics. Performance evaluation was conducted through Pearson and Spearman correlation coefficients to ensure robust and interpretable assessments. The results revealed that GIST-Embedding-v0 consistently achieved the highest average correlation scores across all datasets, indicating strong generalizability. Nevertheless, MUG-B-1.6 demonstrated superior performance on datasets 2, 6, and 7, while UAE-Large-V1 outperformed other models on datasets 3 and 5, thereby underscoring the influence of dataset-specific characteristics on embedding model efficacy. These findings highlight the importance of adopting a dataset-aware approach in embedding model selection for STS tasks, rather than relying on a single universal model. Moreover, the observed performance divergence suggests that embedding architectures may encode semantic relationships differently depending on domain-specific linguistic features. By providing a detailed evaluation of model behavior across varied datasets, this study offers a methodological foundation for embedding selection in downstream NLP applications. The implications of this research extend to the development of more reliable, scalable, and context-sensitive STS systems, where model performance can be optimized based on empirical evidence rather than heuristics. These insights are expected to inform future investigations on embedding adaptation, hybrid model integration, and meta-learning strategies for semantic similarity tasks.

The effects of polycarboxylate superplasticizer (PCE) on the rheological properties and workability of cement-based composites were investigated by testing parameters such as static yield stress, dynamic yield stress, plastic viscosity, slump flow, bleeding rate, and penetration depth. The correlation between the dosage of PCE and the rheological parameters of fresh cement-based composites was analyzed. The results indicated that with an increase in the PCE dosage, the static yield stress, dynamic yield stress, and plastic viscosity of fresh cement-based composites decreased, demonstrating that PCE can improve the rheological properties of these composites. As the PCE dosage increased, the slump flow and bleeding rate of fresh cement-based composites also increased, but the rate of change decreased at higher dosages. Additionally, with an increase in PCE dosage, the penetration depth gradually increased, while the penetration depth difference ($\Delta {H}$) decreased. Furthermore, the compressive strength of cement-based composite cubes slightly decreased with an increase in PCE dosage.

The accurate determination of the conduit water starting time constant ($T_w$) is critical for optimizing hydro turbine performance and dynamic control in hydropower plants. Instead of relying on conventional calculation methods, machine learning (ML) techniques, specifically long short-term memory (LSTM) networks and multilayer perceptron (MLP) models, have been employed to identify $T_w$. The dataset used for model training and validation comprises real operational data collected from two hydropower plants. The effectiveness of both algorithms in $T_w$ identification has been evaluated through simulation, with Python serving as the primary programming environment. The findings indicate that, despite its more complex architecture, LSTM does not necessarily yield superior results. In contrast, MLP, as a relatively simpler model, demonstrates greater accuracy in estimating $T_w$, suggesting that intricate network structures are not always required for precise identification. Additionally, an optimization function ($F_\text{opt}$) has been utilized to assess the reliability of the identified $T_w$ values by comparing them with actual hydro turbine responses. The results underscore the practicality of MLP in hydropower system modeling, providing a computationally efficient alternative for conduit water starting time constant identification. These insights contribute to improving real-time turbine control and enhancing the efficiency of hydropower generation.

Transformer-based language models have demonstrated remarkable success in few-shot text classification; however, their effectiveness is often constrained by challenges such as high intraclass diversity and interclass similarity, which hinder the extraction of discriminative features. To address these limitations, a novel framework, Adaptive Masking Bidirectional Encoder Representations from Transformers with Dynamic Weighted Prototype Module (AMBERT-DWPM), is introduced, incorporating adaptive masking and dynamic weighted prototypical learning to enhance feature representation and classification performance. The standard BERT architecture is refined by integrating an adaptive masking mechanism based on Layered Integrated Gradients (LIG), enabling the model to dynamically emphasize salient text segments and improve feature discrimination. Additionally, a DWPM is designed to assign adaptive weights to support samples, mitigating inaccuracies in prototype construction caused by intraclass variability. Extensive evaluations conducted on six publicly available benchmark datasets demonstrate the superiority of AMBERT-DWPM over existing few-shot classification approaches. Notably, under the 5-shot setting on the DBpedia14 dataset, an accuracy of 0.978±0.004 is achieved, highlighting significant advancements in feature discrimination and generalization capabilities. These findings suggest that AMBERT-DWPM provides an efficient and robust solution for few-shot text classification, particularly in scenarios characterized by limited and complex textual data.

Ensuring the integrity of goods during cold chain transportation remains a critical challenge in logistics, as it is essential to preserve product quality, freshness, and compliance with stringent safety standards. Strategic decision-making in this context requires the prioritization of customer requirements and the optimal allocation of limited operational resources. In response to these demands, an integrated Multi-Criteria Decision-Making (MCDM) model was developed by combining the Best-Worst Method (BWM), Quality Function Deployment (QFD), and Measurement of Alternatives and Ranking according to Compromise Solution (MARCOS) approach. Within this framework, BWM was utilized to determine the relative importance of user requirements, which were then mapped onto specific operational resources through QFD to identify critical resource elements and derive their corresponding weights. These weights, subsequently treated as evaluation criteria in the MARCOS method, were applied to assess the performance of Third-Party Logistics (3PL) providers. The proposed methodology was validated through a case study involving eight user requirements and seven key resources. The findings indicated that precise temperature control and delivery speed were the most critical user requirements, whereas advanced temperature sensors and vehicles with cooling systems were identified as the most significant resources. Based on the MARCOS evaluation, Provider 1 emerged as the most optimal 3PL alternative. This integrated decision-making model offers a systematic and data-driven approach for aligning customer priorities with resource capabilities, thereby enabling logistics providers to enhance service quality, operational efficiency, and strategic competitiveness in temperature-sensitive supply chains. The model also demonstrates practical scalability and adaptability across diverse cold chain scenarios.

The rice crisis represents a significant threat to food security and economic stability in Southeast Asia, a region where rice serves as the primary staple for the majority of the population. This crisis is exacerbated by a confluence of factors, including climate change, crop failures, and restrictive export policies, as exemplified by the El Niño phenomenon and India’s 2023 rice export ban. Rising rice prices have been linked to increased social unrest, with the potential to trigger widespread demonstrations across affected nations. To proactively address this issue, the restlessness indicator was introduced as a predictive tool, integrating key variables such as rice prices, consumption patterns, and per capita income. This study employs a Spatio-Temporal Autoregressive (STAR) model to forecast restlessness values across six Southeast Asian countries—Indonesia, the Philippines, Thailand, Vietnam, Malaysia, and Cambodia—from 2024 to 2028. The STAR (5,1) model was identified as the optimal framework, achieving a Mean Absolute Percentage Error (MAPE) of 15.1%. The forecasting results indicate that none of the analyzed countries are projected to enter a state of unprecedented restlessness during the specified period, suggesting that no severe rice crisis is anticipated within this timeframe. These findings provide critical insights for policymakers and stakeholders, enabling the development of preemptive strategies to mitigate potential food security challenges. The study underscores the utility of the restlessness indicator as a robust tool for monitoring and forecasting rice-related crises, contributing to the broader discourse on sustainable food systems in Southeast Asia.

Efficient bidirectional energy exchange between an alternating current (AC) grid and a direct current (DC) source has been enabled through advanced power converter topologies. In this study, a single-stage AC-DC dual active bridge (DAB) converter employing phase-shift modulation (PSM) was investigated, with a particular focus on performance within the overmodulation regime. Bidirectional switching modules were implemented on the AC side to facilitate seamless energy transfer. Two conventional modulation strategies—sinusoidal and triangular—and a novel back-calculated modulation method were examined for their performance in both linear and overmodulation operating regions. The proposed back-calculation method incorporates an off-line generated reference current waveform designed to approximate linear control characteristics while substantially minimizing current harmonic distortion under overmodulated conditions. This approach extends the linear relationship between the reference current and power transfer capability beyond the conventional modulation limits, thereby enhancing converter performance in high-demand scenarios. Simulation-based analysis demonstrated that, in the linear region, the proposed method reduced average current total harmonic distortion (THD) by at least 45% when compared to conventional sinusoidal and triangular modulation techniques. Moreover, within the overmodulation regime, the linear correlation between the reference current and power transfer was extended by approximately 16.5%. The current harmonic distortion remained below 5% and 8% at modulation ratios of 108% and 112%, respectively, underscoring the robustness of the proposed strategy. These results suggest that the proposed PSM method is highly effective in achieving improved power exchange with reduced harmonic content in both linear and overmodulated operation, thereby offering a viable solution for high-performance AC-DC power conversion in smart grids and renewable energy systems.