The core challenge of college student entrepreneurship risk assessment is that traditional models often ignore complex factor interactions, resulting in biased risk predictions. To address this issue, this paper applies an XGBoost model with SHAP-based interpretation. The implementation includes systematic data preprocessing, including cleaning, encoding, and missing-value imputation, followed by a recursive feature elimination strategy combined with XGBoost feature importance to select the optimal subset of predictive variables. Model hyperparameters are optimized through grid search and cross-validation. Finally, the SHAP framework is used to calculate the marginal contribution of each feature and generate global and local attribution graphs, revealing key drivers of risk prediction. Experimental results confirm the effectiveness of the proposed model. In multi-scenario comprehensive evaluation, the average accuracy exceeds 0.85, the misclassification rate is below 0.15, the F1-score exceeds 0.87, Cohen’s Kappa exceeds 0.71, and crossdisciplinary generalization is strong, with AUC exceeding 0.91. The study demonstrates that the XGBoost-SHAP framework improves the reliability, transparency, and interpretability of entrepreneurship risk assessment under complex nonlinear feature interactions.
To address the limitations of traditional emotional-feature extraction methods for musical performances, including low accuracy, inadequate temporal-continuity modeling, and weak differentiation of subtle emotional changes, this study proposes a multi-channel convolutional attention-based method for extracting emotional feature variations. A multimodal data acquisition system for university music performances is first constructed to collect audio data, including pitch, rhythm, and dynamics, and visual data, including facial expressions and body movements, from vocal and instrumental performances. A multi-channel convolutional neural network is then designed for multidimensional feature extraction from heterogeneous data. The audio channel uses 1D-CNN to capture time-domain and frequencydomain features, while the visual channel combines 2D-CNN and 3D-CNN to extract spatial and temporal sequence features. A hybrid attention mechanism integrating channel attention and temporal attention is introduced to adaptively weight key emotional features and enhance the capture of dynamic emotional changes. Experiments on the self-built UMPED dataset and the public MusicNet dataset evaluate emotion-feature similarity, classification accuracy, and temporal emotion-change fitting, verifying the effectiveness of the proposed multimodal signal-processing framework.
This study systematically explores the application effect of Task-Based Language Teaching in Japanese oral teaching, aiming to solve the problem of disconnection between language learning and practical communication. Using literature research to summarize core theories, and combining questionnaire surveys and classroom observations to analyze current teaching conditions, the study designs a TBLT-based teaching plan suitable for college Japanese oral instruction. A 16-week empirical study is conducted among Japanese majors at a university. The results show that TBLT significantly improves students’ oral accuracy, fluency, and communicative effectiveness, with increases of 15.8 points, 18.6 points, and 19.2 points, respectively. It also effectively stimulates students’ learning interest and willingness to speak. The research indicates that TBLT is highly consistent with the objectives of Japanese oral teaching, and that its effective implementation depends on scientific task design, standardized teaching procedures, and a diversified evaluation system. This study provides an operable reform plan for college Japanese oral teaching and offers theoretical and practical reference for localized TBLT application in small-language education.
Big data-driven precision ideological and political education uses multidimensional student data to meet the demand for innovative talent cultivation in colleges and universities. Based on the theoretical logic of big data technology and personalized ideological and political education, this paper focuses on student-management scenarios and constructs a personalized guidance program featuring “data collection-intelligent analysis-precise delivery-feedback optimization.” By integrating behavioral data, ideological dynamics, academic development data, and psychological information, machine learning algorithms are used to build students’ ideological and behavioral profiles, enabling precise matching of educational content, methods, and timing. Taking 2,000 students from three different types of colleges as research subjects, an empirical study conducted over one academic year shows that the program significantly improves students’ ideological recognition, behavioral standardization, and academic achievement rates. The results verify the feasibility and effectiveness of big data-driven precision ideological and political education and provide a replicable paradigm for scientific transformation of student management in the new era.
As advanced electromagnetic sensing and intelligent signal processing systems increasingly rely on robust feature extraction and multimodal information fusion, modeling heterogeneous cultural information provides useful insights for cross-domain representation learning. This study constructs a hybrid neural network model based on a multi-layer perceptron and self-organizing mapping to simulate the dynamic evolution of identity when individuals encounter heterogeneous musical cultures. The model couples three modules—music feature extraction, cultural distance perception, and self-concept adjustment—while integrating the parallel internalization of musical structures and the symbolic logic of patterns as coupled cognitive processes. Through 500 simulation cycles, the proposed framework reproduces the nonlinear trajectory of identity evolution and identifies three representative stages: cognitive accumulation during initial contact, identity structure reorganization during deep immersion, and pluralistic identity integration. Simulation results reveal an inverted U-shaped relationship between cultural adaptability and learning depth, where moderate exposure promotes integration whereas excessive intensity may trigger emotional ambivalence and identity instability. Sensitivity analysis over 100 parameter sets further demonstrates that cultural openness and music feature extraction accuracy determine statistically significant bifurcation points (p < 0.01). The proposed computational framework provides a quantitative perspective for modeling dynamic identity evolution and offers methodological references for multimodal feature learning and intelligent information processing in complex heterogeneous environments.
Multimodal semantic fragmentation and inaccurate user-intent modeling reduce the effectiveness of music teaching resource recommendation. To address these problems, this paper proposes MMT-MERec, a Multimodal Transformer for Music Educational Recommendation framework that integrates a multimodal Transformer with an educational knowledge graph. The method extracts 768-dimensional semantic vectors from audio, sheet music, video, and text through four pre-trained models: AST, MusicBERT, VideoMAE, and Sentence-BERT. A six-layer Cross-Modal Transformer encoder then performs fine-grained semantic fusion, using audio as a query to align the other modalities. A RotatE-embedded Skill Knowledge Graph containing 142 nodes constrains the recommendation loss to maintain teaching logic. A Skill-Aware Contrastive Learning objective based on user dynamics, including error rate and session duration, is used to optimize resource ranking. On an 8,200-sample dataset, MMT-MERec significantly outperforms the M3oE baseline, achieving HR@5 of 0.603 and NDCG@10 of 0.537. In cold-start scenarios, Cold-Start HR@5 reaches 0.401. These results show that the model jointly optimizes multimodal semantic alignment and educational knowledge constraints, improving recommendation relevance and teaching effectiveness.
As the core aesthetic characteristic of ethnic instrumental art, charm is a highly unified expression of technique and emotion. It embodies the aesthetic pursuit and cultural spirit of Chinese national music. In response to the current problems of emphasizing technique over charm and generalizing emotional communication in ethnic instrumental performance, this article takes representative instruments such as erhu, guzheng, and bamboo flute as research objects and systematically explores the technical implementation path and emotional communication strategy of charm. The study first analyzes its aesthetic connotation and cultural origins, clarifying core characteristics such as the interplay of reality and virtuality, the fusion of emotion and scenery, and the combination of hardness and softness. It then examines how bowing, fingering, and breath techniques support charm, revealing the application logic of glissando, vibrato, breath control, and related performance methods. Although the research remains centered on music performance, its treatment of timbre, breath, and phrase-level signal variation provides a structured reference for broader wave-based expression analysis and human-centered signal interpretation in engineering communication. The study finally constructs an emotional communication system of technical foundation, emotional brewing, and artisticconception sublimation, clarifying the transformation mechanism from technical control to emotional resonance.
With the increasing deployment of intelligent teaching environments and wireless digital infrastructures, efficient multimodal information acquisition and transmission have become essential for objective assessment in vocational education. To overcome the subjectivity and limited dimensionality of conventional evaluation methods, this study proposes a digital literacy evaluation model for young vocational education teachers based on a multimodal Transformer fusion network. Semantic representations of instructional texts are extracted using BERT, spatiotemporal behavioral features from classroom videos are learned through 3D Convolutional Neural Networks, and emotional acoustic characteristics are obtained from speech signals using OpenSMILE and temporal modeling. A cross-modal attention mechanism constrained by the Technological Pedagogical Content Knowledge (TPACK) framework is introduced to integrate heterogeneous information and dynamically optimize feature weighting across instructional scenarios. An end-to-end multi-task prediction architecture subsequently generates quantitative evaluations for six digital literacy dimensions with enhanced interpretability. Experimental results demonstrate approximately 92% classification accuracy on the test set, while quantitative scores for all six secondary dimensions exceed 85.5 in both theoretical instruction and practical training environments. The proposed framework establishes a high-precision and objective assessment methodology for teacher digital literacy and provides an effective engineering solution for multimodal data fusion, intelligent educational sensing, and distributed information processing. Furthermore, its multimodal perception and communication framework offers valuable insights for electromagnetic-enabled smart education systems, wireless sensing platforms, and next-generation intelligent information transmission technologies.
To address the core challenges of internal audit systems in cloud environments, including the trade-off between multisource sensitive data privacy and audit efficiency, rigid process execution, and insufficient mining of hidden associations in distributed enterprises, this study proposes an information-based internal audit decision-making support system integrating lightweight federated learning, adaptive differential privacy, and a BPMN-based automated process engine. Considering the increasing demand for secure intelligent information processing in cloud-edge infrastructures that also underpin large-scale electromagnetic information systems, the proposed framework enables privacy-preserving collaborative analysis while maintaining high computational efficiency. At the edge layer, a four-level privacy classification strategy is employed, where LSTM-based anomaly detection models are trained locally and only ϵ-differential privacy-protected encrypted gradients are uploaded for FedAvg aggregation in the cloud. Rényi differential privacy dynamically adjusts privacy budgets, while entropy weight-TOPSIS risk evaluation and a Neo4j knowledge graph drive adaptive task scheduling through the BPMN engine. NLP-based report generation, Elasticsearch logging, and blockchain anchoring further ensure traceability and reliability. Experimental results demonstrate an average end-to-end latency of 8.4±1.4 s, audit coverage of 93.9±2.2%, risk identification F1-score of 89.2±2.7%, data re-identification rate of 2.5±1.0%, and process automation exceeding 89%. The proposed framework effectively resolves the privacy–efficiency– process dilemma and provides a secure, intelligent, and scalable decision-support paradigm for cloud-enabled enterprise auditing, while offering methodological insights for trustworthy information processing in distributed electromagnetic and cyber-physical infrastructures.
Accurate detection of small-scale targets in UAV aerial imagery remains challenging due to severe scale variation, background interference, and feature degradation caused by repeated downsampling. To address these issues, this study proposes YOLO-SWIFT, a wavelet-integrated feature transformation network designed for small-object detection. First, a Position-Aware Coordinate Downsampling module is developed to preserve critical spatial information during feature compression through coordinate attention and skip connections. Second, a High-Resolution Feature Aggregation Network is introduced to establish an additional high-resolution detection branch for enhanced cross-scale feature fusion. Third, a Wavelet Bottleneck Enhancement module incorporating multi-level wavelet decomposition and a High-Frequency Retention pathway is designed to improve fine-detail representation while expanding the effective receptive field. Finally, an Adaptive Scale-aware Regression IoU loss function is proposed to dynamically balance localization and shape-consistency constraints for small targets. Experimental evaluation on the VisDrone2019 benchmark demonstrates that YOLO-SWIFT achieves 38.7% mAP50 and 22.5% mAP50:95. The proposed framework provides an effective solution for intelligent aerial sensing and offers potential applications in electromagnetic imaging, remote sensing interpretation, and autonomous surveillance systems.
To address the inadequacy of existing models in capturing spatial position information and multi-scale periodic patterns in traffic flow, this paper proposes a Position-Aware Spatio-Temporal Graph Convolutional Network (PASTGCN). The model incorporates a spatio-temporal position embedding module that encodes geographic coordinates and topological attributes into spatial embeddings, combined with sinusoidal encoding for periodic temporal representations. A dual graph learning mechanism fuses a static distance graph based on a Gaussian kernel with an attention-based dynamic graph, while spatial-aware dilated causal convolution enables temporally modulated spatial feature injection. Multigranularity periodic branches further enhance the modeling of daily and weekly patterns. Extensive experiments on the METR-LA dataset show that PASTGCN achieves an RMSE of 2.53 (veh/5 min) and an R2 of 0.982, outperforming several contemporary baselines including DCRNN, STGCN, Graph WaveNet, ASTGCN, AGCRN, and STAEformer. Ablation studies confirm the contribution of each module, with the dynamic graph learning alone reducing RMSE by 0.34. The model also maintains high efficiency with an inference latency of 18.7 ms and a throughput of 53.5 samples/second, demonstrating its suitability for real-time urban traffic prediction. Beyond intelligent transportation applications, the proposed position-aware spatio-temporal modeling strategy provides a practical reference for information fusion and dynamic state estimation in large-scale wireless sensing networks and electromagnetic signal-driven monitoring systems, supporting future intelligent perception and communication infrastructures. This high accuracy and low-latency capability also makes PASTGCN particularly promising for supporting intelligent logistics scheduling in time-sensitive industries.
Against the background of global industrial iteration, collaborative innovation between core enterprises and suppliers has become an important path for improving technological competitiveness and resource-integration efficiency in manufacturing systems. As two major governance mechanisms, trust and contract can reduce cooperation risks, promote knowledge flow, and support joint innovation. Based on transaction cost theory, social exchange theory, and resource dependence theory, this paper constructs a theoretical model of the dual-dimensional effects of trust, including cognitive trust and emotional trust, and contract, including formal contract and relational contract, on collaborative innovation performance. It further examines the mediating role of knowledge sharing and the moderating effect of environmental dynamism. Questionnaire data from 286 core manufacturing enterprises are analyzed using SPSS 26.0 and AMOS 24.0. The results show that cognitive trust, emotional trust, and relational contract significantly improve collaborative innovation performance, while the direct effect of formal contract is not significant. Knowledge sharing partially mediates the effects of cognitive trust, emotional trust, and relational contract, and fully mediates the effect of formal contract. Environmental dynamism positively moderates the relationships between cognitive trust, relational contract, and knowledge sharing. The study provides governance references for collaborative innovation in complex engineering supply chains, including high-reliability electromagnetic equipment manufacturing.
With the rapid development of intelligent communication networks and Electromagnetic Waves, Antennas and Propagation technologies, spatiotemporal information modeling and dynamic topology analysis have become fundamental requirements for large-scale heterogeneous data processing and distributed decision-making. To address the challenge of identifying complex transmission mechanisms in digital inclusive finance, this study proposes an Attention-based Spatio-Temporal Graph Convolutional Network (ASTGCN) that integrates graph convolution, temporal convolution, and multi-head attention for dynamic path recognition. By constructing multidimensional spatiotemporal graphs from multi-source panel data and jointly modeling geographic and economic relationships, the proposed framework captures nonlinear spatial spillover effects and long-term temporal dependencies while enabling interpretable quantification of key transmission paths. Empirical results demonstrate that technological innovation represents the dominant driving channel with a relative contribution of 38.6%, while the proposed ASTGCN model reduces root mean square error and mean absolute percentage error by 18.4% and 21.2%, respectively, compared with the standard LSTM baseline. Furthermore, the model successfully reconstructs regional information propagation topology and reveals dynamic transitions in transmission mechanisms through attention-based path weighting. The proposed architecture provides an effective engineering framework for spatiotemporal graph learning, distributed information propagation, and adaptive network analysis, offering valuable methodological references for intelligent sensing, communication-oriented monitoring, and Electromagnetic Waves, Antennas and Propagation applications.
This research addresses the challenge of automatically identifying complex tennis techniques and tactics, dynamically predicting shot intentions, and assessing round outcomes early from structured event sequences. By constructing a Transformer sequence model, the study provides a data-driven intelligent solution for tactical decision-making. The model uses structured shot sequences from Match Charting Project logs, integrating attributes like shot type, direction, and match status. It employs a six-layer Transformer encoder with a multi-head self-attention mechanism for multi-task learning, jointly optimizing technical/tactical classification, shot intention prediction, and early outcome assessment. Experiments validate its robust performance, achieving a Macro-F1 of 0.935 for tactical classification and a round outcome AUC@5 of 0.923. The model significantly outperforms BiLSTM and CNN-LSTM baselines, with ablation studies confirming the critical contributions of its embedding and multi-task design. The proposed framework enables end-to-end deep analysis and effective prediction of tennis technical and tactical behaviors without visual input, providing a scalable, data-driven solution for intelligent competitive analysis, principles that may also inform structured event-stream prediction in other non-visual engineering datasets.
Uncoordinated charging of large-scale electric vehicles exacerbates peak-valley differences and voltage exceedance risks in the power grid, while existing scheduling methods still have limitations in distributed decision-making, dynamic pricing, and multiobjective balancing. These problems become more significant in charging station clusters where power-electronic converters, communication links, and complex electromagnetic operating environments jointly affect grid interaction stability. In this paper, a collaborative optimization framework based on multi-agent reinforcement learning is proposed for orderly charging at electric vehicle charging stations and coordinated interaction with the power grid. First, each charging station is modeled as an autonomous agent, and distributed environment modeling is realized based on local observation information and Markov decision processes. Second, a proximal policy optimization algorithm is used to generate a dynamic service fee multiplier in a continuous action space, which is combined with a demand elasticity module to form an adaptive pricing mechanism. Finally, a composite reward system integrating grid stability, operational revenue, and user satisfaction is developed, and multi-agent convergence training is achieved through parameter sharing and generalized advantage estimation. The results confirm the overall benefits of joint optimization in load shaping, economic performance, and robustness, providing a technical reference for intelligent charging coordination under grid interaction and electromagnetic compatibility constraints.
Driven by the dual goals of global carbon neutrality and ecological city construction, the integrated application of sustainable landscape materials and recycled fibers has become an important approach to improving urban environmental quality and reducing infrastructure carbon emissions. As intelligent infrastructure increasingly relies on electromagnetic sensing and structural health monitoring technologies for lifecycle assessment, comprehensive material performance evaluation is also essential for reliable engineering applications. However, current landscape engineering still suffers from insufficient compatibility between material properties and ecological requirements, incomplete evaluation systems, and a lack of systematic optimization strategies. To address these issues, this study constructs a multidimensional performance evaluation framework covering mechanical, ecological, durability, and economic performance. Representative sustainable landscape materials, including permeable concrete, recycled stone, ecological wood, and vegetation concrete, are selected as research objects. Through experimental testing, numerical simulation, and case analysis, the key performance indicators and influencing mechanisms are systematically investigated, and performance-oriented optimization strategies are proposed. The results provide theoretical support for sustainable material selection and practical guidance for intelligent infrastructure construction, while offering useful references for electromagnetic-assisted monitoring and long-term performance evaluation in advanced engineering environments.
Driven by the dual-carbon strategic goal, the solar cell industry has developed rapidly, while production efficiency, line balance, and quality control have become increasingly prominent engineering issues. As solar cells convert incident electromagnetic radiation into electrical energy, stable and efficient production-line management is directly related to photovoltaic device performance and clean-energy supply. This paper takes the solar cell production line as the research object and systematically explores an optimization model based on production-line collaboration and process linkage. The development background of the photovoltaic industry and the pain points of traditional production management are reviewed, and the core connotation and theoretical basis of the model are clarified. Implementation paths are proposed from three aspects: cross-process data interconnection, process-cycle matching and bottleneck optimization, and a closed-loop quality traceability system. Practical results show that the model improves the balance rate of the production line, shortens the production cycle, reduces process waiting waste, and supports the transformation of solar cell manufacturing toward digitalization and intelligence. The research provides a reference for high-quality development of photovoltaic and electromagnetic-energy conversion industries.
Current evaluations of traditional Chinese medicine intervention for diabetic nephropathy often lack data-driven predictive tools, resulting in limited objectivity and inconsistent outcome assessment. To construct an objective efficacy prediction method, this study included clinical data from 2,156 patients and extracted key variables through data preprocessing and feature engineering. Logistic regression, random forest, support vector machine, and XGBoost were used for model training and comparison. The results show that the XGBoost model achieved the best predictive performance, with an AUC of 0.902. Interpretability analysis indicates that baseline urinary protein, estimated glomerular filtration rate, and traditional Chinese medicine blood stasis syndrome are important predictive factors. Subgroup analysis shows that early-stage patients without significant blood stasis syndrome obtained the best efficacy, with an effectiveness rate exceeding 80%. By integrating artificial intelligence with clinical data analysis, this study demonstrates the feasibility of quantifying traditional Chinese medicine efficacy and provides support for personalized treatment evaluation. The modeling strategy also offers references for biomedical signal analysis, intelligent diagnosis, and data-driven sensing applications.
Blended practical training in engineering electromagnetic measurement is becoming increasingly prevalent in the context of digital transformation in education. However, due to the inherent complexity of this field, traditional evaluation methods are inadequate for effectively quantifying its technical challenges and accurately assessing student learning outcomes. To address this issue, this study proposes a blended practical training learning effectiveness evaluation method based on a combined EWM-AHP-DEMATEL weighting model. The method employs the Analytic Hierarchy Process (AHP) to determine subjective weights, the Decision Making Trial and Evaluation Laboratory (DEMATEL) technique to analyze inter-indicator correlations and dynamically adjust weights, and the Entropy Weight Method (EWM) to quantify objective weights from the information entropy of measured data, thereby establishing a comprehensive subjective-objective weighting framework. Empirical analysis of representative engineering electromagnetic measurement applications demonstrates that the proposed model effectively preserves expert experiential judgment while significantly reducing subjective bias through objective data-driven entropy analysis. This study provides a scientific diagnostic tool for optimizing blended instruction in electromagnetic measurement training and also offers valuable reference for cultivating high-quality applied talents in the field of electromagnetic non-destructive testing and sensing.
Museum exhibition design often suffers from fragmented narrative logic and one-way information interaction, leading to a mismatch between exhibition narrative value and audience engagement depth. To address this issue, this paper constructs a narrative exhibition design framework from the perspective of interaction design. By establishing an interaction model of information flow, behavior flow, and emotional flow, the study proposes a system optimization path based on user behavior data and spatial feedback loops. Eye-tracking and behavior recording technologies are used to collect browsing paths and dwell times from 120 visitors in different exhibition areas, forming an interactive behavior database. A multidimensional interaction matrix is then established based on content narrative nodes and audience behavior responses, and cluster analysis and weight allocation are applied to optimize the narrative structure. Prototype exhibition interface testing is conducted, and facial recognition and questionnaires are used to evaluate changes in audience emotion. The results show that the optimized exhibition model increases average visitor dwell time by 32.5% and interaction triggers by 40.7%, verifying the effectiveness of interaction design in narrative expression and participation enhancement.
