朝陽科技大學

Background Dynamic binding capacity at 5% breakthrough (DBC5%) is a key performance parameter in packed-bed chromatography because it determines the effective working capacity of the adsorbent and directly influences process productivity under dynamic-flow conditions. Methods This study presents a systematic methodology for the predictive optimization and scale-up of DBC5% for C-phycocyanin (CPC) purification from Spirulina platensis using a sequential design of experiments (DoE) approach. A 2⁴ full factorial design with two center points (2⁴ + 2) was first employed to identify the significant operating variables, followed by a central composite design (CCD) to evaluate potential nonlinear responses. Significant findings The 2⁴ + 2 factorial model exhibited superior predictive performance, achieving an R² of 98.97% and a predicted R² of 94.82%, compared with corresponding values of 91.04% and 48.50%, respectively, for the CCD model. The optimized operating conditions (pH 6.0, 10% (w/v) feed concentration, 1.6 cm bed height, and a flow rate of 10.0 mL/min) yielded a predicted DBC5% of 10.51 mg/mL, which was experimentally validated by an observed value of 10.45 mg/mL, corresponding to a relative error of 2.9%. Furthermore, scale-up from 1.6 cm to 5.0 cm internal-diameter columns while maintaining hydrodynamic similarity successfully preserved DBC5%, demonstrating consistent adsorption performance across the investigated scales. Although the developed regression model is specific to the chromatographic system investigated, the proposed DoE-based optimization framework and hydrodynamic scale-up strategy provide a practical methodology that can be applied to other packed-bed chromatography systems following appropriate experimental calibration and validation.

Why does the same unmanned retail technology produce different emotional responses across formats? Integrating the Stimulus-Organism-Response framework with Cognitive Appraisal Theory and the Technology Readiness Index 2.0, we test a dual-pathway model in which automation stimuli are associated with parallel emotional burden (via threat appraisal) and emotional uplift (via benefit appraisal), with retail format positioned as a categorical moderator capturing format-level automation match. On-site intercept data from 483 consumers across three unmanned formats in Taiwan (convenience stores, laundromats, unstaffed gyms) were analyzed using PLS-SEM, NCA, and fsQCA. Positive design features predict benefit appraisals and uplift but do not attenuate threat appraisals, which are primarily associated with perceived human absence. Consistent with our format-moderation hypotheses (H8a, H8b, and H8c), only the threat pathway is significantly moderated by retail format, being strongest in convenience stores and weakest in laundromats; the benefit pathway is largely format-invariant. Benefit appraisal and uplift are necessary conditions for behavioral loyalty, whereas the mere absence of burden is insufficient.

This study aims to improve the prediction of borrower default risk in the peer-to-peer (P2P) lending sector by integrating machine learning techniques with feature selection strategies. Recursive Feature Elimination (RFE) is applied to enhance model transparency and predictive efficiency, addressing both computational and decision-related aspects of credit risk analysis. Using a real-world Lending Club dataset comprising 725,096 loan records, five machine learning models—Random Forest, Logistic Regression, Extreme Gradient Boosting, Multi-layer Perceptron, and K-Nearest Neighbors—were developed and fine-tuned via GridSearchCV. Model performance was assessed using accuracy, AUC, precision, recall, and F1-score. Among these, the Random Forest algorithm achieved the best performance with 91

Classifying code snippet-based questions is essential for teaching, preparing assessment materials, and supporting intelligent learning systems in programming education. Traditional frequency-based encodings, such as TF-IDF, often fail to capture the contextual semantics within code-related questions. This study employs contextualized embeddings generated by the large language model Text-Embedding-3-Large (TE3L) to evaluate their effectiveness in classifying code-related questions. It further investigates which classifier architecture best complements the TE3L representation. Using a small-scale dataset of 171 SQL certification-style questions representative of course-level repositories, we analyze the classification complexity reduced by the TE3L scheme compared to TF-IDF. Then, we investigate classification performance under various classifier architectures with TE3L embeddings, including single models, boosting, and stacking ensembles. Results demonstrate that the TE3L scheme significantly reduces classification complexity and improves performance compared to the TF-IDF. Single classifiers, particularly the support vector machine with a linear kernel and the stochastic gradient descent classifiers, performed the best with the TE3L scheme and achieved an 11-percentage-point relative improvement over the benchmark in the weighted macro-average F1 score. The boosting and stacking techniques did not enhance performance, reflecting the challenges of ensemble learning under small-sample, imbalanced conditions. This work highlights the practical value of using LLM-based embeddings to automate question classification in low-resource educational contexts, supporting teachers in building intelligent assessment tools without requiring deep expertise in NLP or machine learning.

As a powerful detection technique, the surface-enhanced Raman scattering (SERS) has gained significant attention owing to its ability to reveal unique fingerprint information. Especially, the flexible SERS substrate due to its exceptional features such as portability, ease of integration of nanomaterials, rapid in-situ and on-site detection makes it an ideal platform for the real-time detection. This paper proposes a flexible SERS substrate based on carrot cellulose nanofibrils (CCNFs) modified with polyetheramine (M2070) via the freeze-drying technique followed by the photochemical decoration of gold nanoparticles (AuNPs). The amphiphilic structure of M2070 promotes the abundant chelation sites, facilitating the uniform growth and strong adherence of AuNPs throughout the CCNF-M2070 matrix. The fabricated flexible AuNPs@CCNF-M2070 SERS substrate exhibit superior Raman enhancement, low limit of detection of 1.08 × 10–10 M, excellent mechanical durability for over 100 cycles of bending and twisting test, high homogeneity, and reproducibility towards the detection of pesticide, thiram with a relative standard deviation value of less than 10