This study presents a dynamic route-planning framework implemented in Python based on an Elevated Beluga Whale Optimized Feed-Forward Backpropagation Neural Network (EBWO-FFBPNN). The dynamic route-planning framework optimizes the tourism experience by personalizing and dynamically re-adjusting route planning by utilizing a comprehensive data set. Data preprocessing involves normalization and data cleaning to ensure the quality of the input. Feature extraction is completed via Word2Vec for the semantic representation of the textual data, and topics are profiled using Latent Dirichlet Allocation (LDA) for travel and tourist interest. The core FFBPNN model seeks user behavior patterns and predicts route preference optimality. Real-time dynamic adjustment is achieved through EBWO-based dynamic optimization that adapts the route in real time according to the environmental context and user behavior. Experimental results demonstrate a precision of 0.98 and an F1 score of 0.96, indicating superior prediction accuracy, route efficiency, and personalization compared to conventional methods.
The rapid growth of Cross-Border Electronic Commerce (CBEC) has increased demand for personalized English-learning product recommendations. However, traditional recommendation systems struggle to capture complex learner interaction behavior and learning preferences. This study proposes a Dynamic Sand Cat Swarm Attention-based LSTM (DSCS-Att LSTM) model for intelligent recommendation. The model is evaluated using the Amazon Books Reviews dataset, with preprocessing including missing-value imputation and min–max normalization. TF-IDF is used for textual feature extraction, while user interactions are modeled sequentially. The framework integrates attention mechanisms to capture key learning patterns and employs Dynamic Sand Cat Swarm optimization for hyperparameter tuning. Experimental results show high performance with 98.87% accuracy, 98.25% F1-score, and 0.998 AUC, outperforming existing models. The proposed framework analyzes collaborative user interactions and behavioral engagement patterns within CBEC learning environments to improve collaborative recommendation quality.
AI reshapes e-collaboration in entrepreneurship education, yet visible tools are often mistaken for drivers. This study deconstructs the AI-empowered training ecosystem for solo entrepreneurs. A four-stage mixed-methods framework (AHP-DEMATEL-ISM-Monte Carlo) integrates 18 expert judgments with three datasets across 21M+ records, assessed via 10,000 iterations. A 78.48/21.52 asymmetry emerges: 78.48% of systemic importance resides in Layer 2 (human capabilities and pedagogy), while Layer 1 (AI tools and VR) accounts for 21.52%. Cross-boundary collaboration (S5) is the supreme leverage point (weight=0.095, causality=+0.647). Visible tools and VR (S10) show strong negative causality (−0.753). Eight Layer-2 drivers exhibit high robustness (CV < 0.045). This study provides the first robustness-validated quantification of hierarchical asymmetry in collaborative technology adoption. A “capability-first” governance model addresses risks via stakeholder-specific pathways. The 78/22 architecture offers an evidence-based heuristic for optimizing AI-empowered collaborative ecosystems.
Karate training relies on subjective manual evaluation of intricate motions, hindering scalable, consistent coaching. This paper proposes MMAF-Net, a multimodal AI framework for real-time karate action recognition and automated performance feedback. Its three-branch deep learning architecture integrates visual, pose-estimation and inertial sensor streams to extract complementary motion features, fused via a temporal attention module to classify 23 karate action types accurately. Trained and validated on MS-KARD (2.8 million video frames and 5.6 million sensor readings from dual cameras and three IMUs), the model generates explainable coaching tips through rule-based modules referencing prediction confidence, posture bias and motion stability. Tests yield 96.3% accuracy and 95.1% F1-score, outperforming benchmarks like KarateNet. With only 24 ms inference latency, this real-time system suits interactive martial arts training scenarios.
The development of interdisciplinary academic leaders is crucial for universities aiming to enhance collaboration, innovation, and knowledge integration across disciplines. Research introduces a deep learning-based framework, Wingsuit Flying Search-driven Dynamic Recursive Neural Networks (WFS-DRNN), implemented using Python (3.11), to predict leadership potential and optimize development pathways. By modeling hierarchical and sequential relationships among faculty qualifications, research experience, collaborative networks, and mentorship records, the WFS-DRNN captures both explicit and latent patterns critical to leadership growth. The Wingsuit Flying Search Algorithm (WFSA) efficiently explores optimal development pathways, enabling tailored interventions, such as interdisciplinary project participation, mentorship opportunities, and skill enhancement programs. Experimental results indicate that WFS-DRNN surpasses traditional heuristic and statistical methods, achieving 95.8% accuracy, 95.2% precision, 92.75% recall, and a 95% F1-score.
Innovation and entrepreneurship education requires objective evaluation, but existing methods rely on subjective judgments, single-source data, and fragmented indicators, limiting their reliability and scalability. This study proposes an Artificial Intelligence-driven framework using multisource information fusion and a weighted group search–bidirectional long short-term memory network (WeightedGS-BiLSTM-Net). The model was tested on a dataset of 214,354 records including academic, innovation, entrepreneurial, and textual feedback data. Structured data were normalized, while text was processed and embedded using bidirectional encoder representations from transformers. Principal component analysis reduced dimensionality and Dempster–Shafer theory fused features. The proposed model achieved 98.67% accuracy, enabling scalable and reliable evaluation of innovation and entrepreneurship.
This research proposes a comprehensive framework for texture generation and composition optimization of Chinese paintings based on fuzzy rules. A Discrete New Caledonian Crow Learning-mutated Adaptive Neuro-Fuzzy Inference System (DNCCL-ANFIS) is employed to guide the texture generation process. A feedback loop incorporating evaluation metrics, including 0.95 Structural Similarity Index (SSIM), 98.6% accuracy, 35.5 dB Peak Signal-to-Noise Ratio (PSNR), and aesthetic scoring, ensures iterative refinement of both textures and layout. Comparative evaluations against Variational Autoencoder (VAE), Variational Autoencoder with Particle Swarm Optimization (VAE+PSO), Enhanced Sparrow Search Algorithm–Enhanced Gated Recurrent Unit (ESSA-EGRU), Generative Adversarial Network (GAN), and Convolutional Neural Network (CNN)-based methods demonstrate the superiority of the proposed framework, which successfully generates realistic textures and optimized compositions while preserving traditional Chinese artistic principles.
The real-time feedback systems enable instant corrections and address organizational issues such as delays, poor accuracy, and inflexibility in conventional sports monitoring systems. To mitigate these shortcomings, the research explores the application of Deep Reinforcement Learning (DRL) in real-time sports training, focusing on dynamic feedback and performance modulation for volleyball players. This research presents an approach that uses deep reinforcement learning for providing feedback on athletic training sessions. The presented method takes into account such issues related to existing monitoring devices as latency, poor accuracy, and inflexibility. The deep reinforcement learning technique is used to provide feedback on athletes' performances during the volleyball game. Physiological and kinematics data is pre-processed using min-max normalization and Kalman filtering. Experiments indicate better performance compared to conventional methods based on deep reinforcement learning techniques.
Aiming at the problems of complex multi-agent collaboration, insufficient process transparency, and unclear responsibility attribution in scientific and technological translation projects, this paper introduces blockchain technology to construct the whole process management framework of scientific and technological translation projects. By chaining the key links such as task allocation, process tracking, responsibility definition, and performance evaluation, the traceability and automation management of the project process are realized. The empirical analysis based on real project data shows that the blockchain mechanism has a significant effect in shortening the project cycle, reducing rework rate, and improving data integrity. The research results show that the blockchain provides a feasible path for the digital and intelligent upgrading of science and technology translation project management by reconstructing the cooperation and governance mechanism.
With the widespread application of AI in sports training, this paper proposes a hybrid LSTM-CNN system for real-time martial arts technique estimation and feedback. The framework fuses CNNs for spatial feature extraction and BiLSTMs for temporal movement analysis, enabling accurate recognition of dynamic martial arts motions and providing practitioners with visual feedback on motion deviations to improve form and accuracy. Prototype tests used video data of core martial arts skills under varying movement speeds and lighting conditions, yielding a high recognition accuracy of 94.2% and a low average inference latency of 20 ms, validating real-time operation. A 7-question Likert-scale questionnaire confirmed strong user satisfaction with the system's accuracy, responsiveness, and usability. This model does not replace traditional martial arts teaching but complements it with objective, data-driven, real-time feedback, serving as a training aid to boost learning efficiency and performance via measurable visual and temporal analysis.
The integration of Artificial Intelligence (AI) with interactive art installations presents challenges in achieving real-time responsiveness under limited computational and power resources. This study proposes Mobilized-EfficientNet for Deep Embedding (MENDE), a lightweight deep learning framework designed for embedded environments. The approach utilizes multimodal sensor data, including visual, audio, and motion inputs, which are preprocessed through normalization, resizing, and augmentation to improve robustness. The MENDE architecture is optimized for low-power embedded platforms, ensuring efficient inference with minimal latency. The model is developed and deployed using Python 3.10 on microcontroller-based systems with integrated sensors and real-time control capabilities. Experimental results demonstrate that the proposed system achieves over 97.25% accuracy while maintaining efficient real-time performance. This research provides a scalable and adaptable solution for AI-driven interactive art installations operating in resource-constrained environments.
Sand art features unique visual textures and symbolic narratives yet lacks automated aesthetic assessment. This work presents SASA-CNN, a dual-branch deep learning model for sand art aesthetic evaluation. It integrates ResNet-50 for visual aesthetics and VGG-16 for semantic features, with an attention-guided module fusing dual outputs for regression and classification. Built on transfer learning, the model is pre-trained on public datasets and fine-tuned on 1,247 screened sand art images. Multiple metrics verify its superior performance over baselines. Results prove combining visual-semantic features and attention fusion effectively enables reliable computational aesthetic evaluation for sand art.
Digital/hybrid work makes e-collaboration skills key graduate employability indicators, yet current assessment methods (self-reports, crude skill lists) are ineffective for tech-mediated work. This paper proposes a taxonomy-based approach to quantify such skills from digital traces and analyze their employment correlations. Using the Kaggle Employment Skills dataset, a formal e-collaboration taxonomy (platforms, coordination, artifact collaboration, documentation, workflow integration) was developed. Text processing and rule-based mapping generated skill indicators and an E-Collaboration Skill Index, with predictive models tested for discrimination/calibration. Workflow integration, coordination, and documentation emerged as key signals; the index and its models outperformed baselines. Organized digital footprints effectively measure e-collaboration skills, which correlate significantly with employment outcomes, informing curriculum design, career services, and hiring analytics.
Resource allocation for cultural and creative digital platforms is complex because project evaluation relies on uncertain, multidimensional, and subjective values that are not determined solely by financial performance. This paper proposes a framework, the Fuzzy Creative-Platform Resource Allocation (Fuzzy-CCRA), to prioritize creative projects under uncertainty and make the process transparent and explainable. The framework combines Kickstarter data (648,400 projects) with Behance engagement data to enrich cross-platform analysis. Nine decision criteria were defined and represented using a five-level fuzzy linguistic scale with triangular membership functions. The main approaches to ranking were Fuzzy TOPSIS and Fuzzy VIKOR, and sensitivity analysis was used to assess robustness. The framework produced consistent allocation levels, with Spearman rank correlations above 0.876 in eight scenarios and a Precision@50 of 0.824. Fuzzy-CCRA proposes a powerful, interpretive, and data-driven system of fair distribution of resources on creative platforms.
In China, internet celebrities are becoming increasingly prominent which transforms the economic form including cultural theme. Ensuring customer loyalty, however, remains crucial for profitability. This study analyzes how consumer loyalty toward cultural influencers is generated in live streaming commerce. The elements include influencer image, brand awareness, perceived quality, perceived value towards affective response. The questionnaires collected online were computed. The total meaningful amount was 510 consumers who have visited the cultural theme live streaming online in TikTok. The data were calculated using SEM-ANN analysis. The output shows that all the elements with cognitive have significant effects on affective response which directly impact satisfaction and loyalty. Moreover, ANN showed the importance of different cognitive elements. The results gave a comprehensive understanding of the outcoming associated with loyalty of consumers. And the research contributes to the literature on consumer loyalty to live streaming online cultural influencers in China.
The widespread adoption of online and blended learning has reshaped higher education, yet student engagement frequently diminishes over time, and traditional gamification suffers from the unsustainable novelty effect. This study constructs a Multidimensional Dynamic Incentive Model (MDIM) by synthesizing behavioral incentive theory and self-determination theory, featuring progressive task difficulty, real-time data feedback, and contextualized narrative support. A 32-week longitudinal quasi-experiment was carried out with two parallel classes, using mixed-effects and cross-lagged models for data analysis. Results indicate that DDIM significantly alleviates mid-term engagement decline, sustains long-term learning participation, and stably enhances academic performance without exacerbating the Matthew effect. This research provides an adaptive, sustainable gamification design for e-collaboration systems and technology-enhanced course development, offering empirical and practical implications for digital learning governance in higher education.
With the rapid development of China's society and economy, small and medium-sized enterprises (SMEs) hold a significant position among Chinese enterprises. The financial management capacity evaluation in SMEs constitutes a quintessential multi-attribute decision-making (MADM) challenge. In recent studies, scholars have increasingly combined methods such as the Logarithmic TODIM (LogTODIM) method and entropy weight to tackle complicated evaluation tasks, while Z-numbers are widely adopted to effectively describe uncertain information with varying degrees of credibility. This paper establishes a Z-number-based LogTODIM (ZN-LogTODIM) model, which is specially designed to handle MADM problems under Z-number environments. To verify the feasibility and efficiency of the proposed approach, a concrete numerical example concerning SMEs' financial management capacity evaluation is conducted. The results show that this method performs well and can be further applied to practical decision-making scenarios in operational and educational management.
With the rise of short video platforms, algorithmic recommendation systems, while enabling personalized content distribution, have been criticized for fueling social media opinion polarization. This study empirically explores how algorithms drive such polarization on mainstream short video platforms. Integrating user behavior, content distribution patterns, and interactive networks, it examines algorithms' filtering and amplification in information flow. Through a theoretical model and multi-source data, it reveals how algorithms shape viewpoint exposure and extreme opinion spread, identifying varying polarization impacts across algorithms and platforms. Via flowcharts and data models, it elaborates on coupling mechanisms among content push, interaction, and differentiation. Finally, targeted algorithm optimization suggestions reduce polarization risks, aiding governance and healthy online ecosystems.
Under the "dual-carbon" goal, hidden emissions in urban construction archives management demand urgent attention. Evidence from Nanjing's 2023 digitalization project reveals a 25% emission rise during transition due to procuring 200+ scanners and 5 servers, highlighting a carbon efficiency gap. Using measured data and scenario simulation, this study uncovers a two-stage "transition-stable" carbon mechanism, establishes a life-cycle accounting framework (formation, storage, inspection, destruction), and proposes a technology-management-policy reduction strategy. Emissions are projected to drop 76.4% by 2025, shifting from paper/transport to electronic storage. Management optimization drives over 50% of savings; a cross-department digital platform alone cuts 3.3 tons CO2e at & YEN;320/ton. Higher standardization enables the Planning Bureau to cut emissions by 65.4%, versus 46.4% for the Urban Management Bureau reliant on on-site access. The study offers practical tools to balance digital transition and carbon control, aiding governance of hidden urban emissions.
This study introduces a knowledge management model integrating deep learning into entrepreneurship education to boost competency development and institutional decision-making. It frames AI-supported education as an organizational learning system that converts learner activities, performance metrics, and project evidence into actionable knowledge for curriculum enhancement and mentorship. The research synthesizes perspectives from educators, students, AI experts, and policymakers, presenting pre/post-AI outcome indicators alongside a four-university cross-institutional analysis. Findings reveal improved learning performance and early entrepreneurship gains, with trained faculty, funding, and industry partnerships proving critical to scalability. The paper outlines a competency mapping framework and feedback-driven workflow, stressing explainability, fairness, equity, and digital divide reduction as prerequisites for responsible AI deployment.
