This study examines the visible thematic patterns of artificial intelligence (AI)–related research in Mainland China, Taiwan, Hong Kong, and Macao from 2021 to 2025. The analysis is based on Scopus-indexed journal articles and uses keyword frequency counts, VOSviewer density visualizations, and KH Coder co-occurrence networks. These methods are applied to describe how AI-related keywords appear across the four regions and how their distributions change during the five-year period. Across all datasets, terms such as machine learning, deep learning, and neural network appear frequently and occupy central positions in the visual outputs. The VOSviewer heatmaps show that regions with larger publication volumes display wider areas of keyword density, while regions with smaller datasets present more compact clusters. Beginning in 2024 and 2025, generative AI–related terms, including large language model and ChatGPT, become visible across all regions. The KH Coder networks illustrate that the four regions contain multiple clusters of co-occurring keywords, with differences in cluster size and distribution reflecting the underlying dataset scale and the topics present in each regional corpus. Overall, the results offer a descriptive account of how AI-related terms appear in the collected datasets and how their visible distributions vary among the four regions during the study period. The findings are intended to summarize observable patterns without inferring causal explanations or evaluating the significance of regional differences.

• View PDF

Metaphor translation plays a key role in cross-cultural communication. Among metaphor types, animal metaphors stand out for their rich cultural connotations and cognitive complexity, making them a valuable testing ground for translation strategies. Despite growing interest, existing research has yet to fully clarify the cultural adaptation mechanisms involved in rendering animal metaphor translation across languages. In particular, how different translation agents dynamically process these culturally loaded expressions remains underexplored, which complicates efforts to optimize human-machine collaboration. This study adopts conceptual metaphor theory and an integrated methodology combining qualitative and quantitative analysis with theoretical interpretation. Drawing on three English translations of Mo Yan’s Frog—by Howard Goldblatt, ChatGPT-4.0, and ChatGLM—this study conducts a systematic comparison of how human and machine translators handle animal metaphors. The analysis shows that effective rendering requires more than literal transfer: it depends on activating culture-specific frames and maintaining evaluative stance, not merely preserving surface imagery. While recent advances in artificial intelligence yield relatively high rates of literal retention, machine translations tend to remain surface-bound when metaphors are culturally or politically charged. By contrast, the cultural awareness and interpretive craft evident in the human translation more consistently preserve metaphorical nuance and ideological force. This study offers new evidence for research on metaphor translation and provides practical guidance for improving human-machine collaborations in literary contexts—e.g., using machine outputs to secure surface mapping while human translators recalibrate cultural frames and stance.

• View PDF

Cancellations on the Dutch Railway network are a common and unpredictable occurrence; however, little research has focused on predicting these cancellations. Previous studies on the Dutch railway system have primarily concentrated on delay prediction. For this regression task, models such as XGBoost, Random Forest, Long Short-Term Memory (LSTM), and Gradient Boosting Decision Tree have been shown to perform well. Graph neural network-based models have been used for regression tasks on other transportation networks. We propose a Dynamic Graph Neural Network (DGNN) combined with an LSTM network for binary classification of cancelled trajectories. We compare the model with baseline models on a seasonal split to compare the feature importance across different seasons. Model performance is gauged using paired t-tests on bootstrapped F1 scores. Additionally, Precision, Recall, Balanced Accuracy, and AUC are considered metrics for further comparison. The newly proposed features achieve mostly positive feature importance scores across the models. Amongst the evaluated models, the proposed DGNN and XGBoost outperform the baseline models. Overall, the models underperform with F1 scores no higher than 0.4. This paper provides insight into the influence of various weather and operational features on cancellations on the Dutch railway network, with the operational features proving insightful.

• View PDF

Deep learning technologies have revolutionised pedagogical techniques in recent years by enabling individualised, adaptive learning environments in English for Specific Purposes (ESP) training. The efficacy of these AI-driven systems depends on how well they align with students' cognitive learning styles, including visual, introspective, and kinesthetic styles, which influence how they process and interact with information. This study examines the impact of cognitive learning styles on student performance and perceptions in deep learning-based ESP classes. Through stratified random sampling, 240 undergraduate students from Universitas Muhammadiyah Gresik participated in the study, which used a mixed-methods explanatory sequential design. Validated tools to evaluate cognitive styles and ESP performance were used to collect quantitative data, while semi-structured interviews with a purposive subsample provided qualitative data. Visual learners performed significantly better than their reflective and kinesthetic peers, as indicated by structural equation modeling (β = 0.42, p < 0.001). The results of a qualitative study showed that visual learners preferred graphical input, reflective learners needed depth and timing, and kinesthetic learners expressed disengagement from static interfaces. Emotional responses, including anxiety and a decline in self-efficacy, emerged as a recurrent pattern among non-visual learners. The study concludes that cognitive congruence has a critical role in determining affective participation and academic success in AI-mediated ESP situations. By emphasising the need for inclusive instructional design that considers a range of cognitive profiles, these discoveries contribute to the discussion of customised learning in digitally enhanced language training.

• View PDF

This study examines how Artificial Intelligence (AI) can leverage the concept of Redefinition—the highest level of the Substitution-Augmentation-Modification-Redefinition (SAMR) model—to transform the teaching of English as a Foreign Language (EFL) reading from comprehension-focused activities to inquiry-driven and effective reading and writing activities. Based on socio-cultural theory, Project-Based Learning (PBL), and theoretical frameworks of digital competence, this study conceptualizes AI as a cognitive mediating framework that supports knowledge transformation rather than task automation. Using a multi-method action research design, interventions were conducted with 130 Vietnamese university students enrolled in intermediate-level English reading courses over eight weeks. Traditional reading activities were redesigned into AI-assisted projects, including multimedia infographics, podcasts, and investigative reading tasks. Quantitative data were collected through a validated 40-item questionnaire, and qualitative insights were gathered from semistructured interviews with six purposefully selected participants. The study results indicate that AI-assisted redefined reading tasks promoted higher-order thinking, synthesis, and practical application, demonstrating the successful implementation of the SAMR Redefinition methodology. Students perceive AI as a supportive learning partner, enhancing comprehension, reducing frustration, and increasing motivation and confidence. A fairly strong positive correlation (r = 0.62) was found between purposeful AI use and student engagement and reading performance. Qualitative results further suggest that AI supports continuous reading, collaborative preparation, and multimodal knowledge building, while encouraging responsible and critical use of AI. The study proposes an AI-powered SAMR framework and a pedagogical sequence of read-interpret-transform-create, providing a scalable model for teaching English as a transformative foreign language reading.

• View PDF

This study focuses on the extraction of ethnic music cultural genes and green communication pathways from a digital humanities perspective. Addressing the digital transformation challenges faced by traditional ethnic music preservation, the study constructed a multimodal cultural gene extraction framework integrating deep learning, natural language processing, and computer vision, extracting 687-dimensional cultural features from 12 types of ethnic music, with the model achieving an F1 score of 0.863. A green communication system based on a cloud-edge-device collaborative architecture with 291 nodes was designed, achieving an energy efficiency ratio of 36,300 people per kilowatt, representing a 62.8% improvement over traditional architectures and an annual carbon emission reduction of 1076.8 t. A real-time energy consumption monitoring and carbon emission accounting system covering six major scenarios was established, with mobile-end optimization rates reaching 52.3%, translating green communication into quantifiable indicators. Systematic solutions were proposed for technical challenges such as sample imbalance and high-dimensional sparsity, increasing data availability to 91.2% and system availability to 99.7%. The successful implementation of 12 projects validated the feasibility of translating theory into practice. The research outcomes provide a computable methodological paradigm for the digital preservation of ethnic music, with potential for extension to digital museums, online education, and other fields, contributing a Chinese solution to the green transformation of the global digital cultural industry.

• View PDF

This study develops an annotated and validated multicultural sentiment dataset derived from Indonesian digital literature. The study integrates Cultural Sentiment Analysis (CSA) and Critical Discourse Analysis (CDA) to address a significant gap in existing research. No prior corpus has systematically combined cultural affective dimensions with linguistic and ethnographic validation, making this dataset crucial for mapping cultural value representations in a multicultural context. The corpus includes over 100 digital literary texts—short stories, online novels, and poems—sourced from platforms such as Wattpad, KBM App, and scholarly blogs, selected through purposive sampling to ensure diverse ethnic and thematic coverage. Annotation was carried out by trained annotators using a culturally grounded emotion lexicon, identifying sentiment polarity (positive, negative, and neutral), cultural values (social harmony, cooperation, spirituality, resistance, and adaptation), and linguistic indicators. Validation involved linguistic review for semantic accuracy and ethnographic verification through Focus Group Discussions with cultural experts from various ethnic groups. The resulting multi-layered dataset provides authentic, contextually grounded, and bias-mitigated representations of cultural sentiment in Indonesian digital literature. Beyond enriching digital humanities scholarship, it offers a reusable open resource for future research, automated sentiment analysis development, and data-driven policy formulation, all aimed at enhancing digital cultural literacy and intercultural understanding in Indonesia.

• View PDF

To address the significant decline in face recognition performance caused by low resolution, high noise, and complex degradation factors in security surveillance scenarios, this paper proposes a joint optimization framework that integrates a Transformer and a Generative Adversarial Network (GAN). The innovation of this framework lies in: (1) designing the Face Reconstruction Transformer (FRFormer), which integrates a hierarchical window attention mechanism and a multi-level feature pyramid structure, enhancing the ability to retain identity features through local-global collaborative modeling; (2) constructing the GFP-GAN reconstruction model, which combines pre-trained face priors and degradation removal modules, and utilizes adversarial training to improve image authenticity and detail restoration. Experiments show that when the input is 32 × 32 pixels, the PSNR of GFP-GAN is increased by more than 8 dB, and the SSIM reaches 0.953; FRFormer achieves recognition accuracies of 99.58% and 96.31% on the LFW and AgeDB-30 benchmarks, respectively, which are 0.08 and 0.13 percentage points higher than those of Swin Transformer. Ablation experiments verify the effectiveness of the window attention mechanism and hierarchical reconstruction strategy, especially in noise suppression and cross-pose recognition tasks. This framework has broad application potential in degraded visual conditions, such as biometric recognition and medical image analysis, and provides an end-to-end solution for low-quality face recognition.

• View PDF

Wireless Sensor Networks (WSNs) are characterized by severe energy constraints, dynamic topology, and limited computational resources, making routing design a critical challenge. Traditional single-path and static routing protocols often lead to uneven energy consumption and premature node failures, thereby reducing network lifetime. To address these limitations, this paper proposes an energy-aware multipath routing protocol that integrates AntNet with a lightweight Multilayer Perceptron (MLP) model. Unlike existing artificial neural network-ant colony optimization (ANN-ACO) or deep learning based routing approaches, the proposed method does not embed complex learning mechanisms into the routing core. Instead, the MLP model is used as an auxiliary decision-support component to assist AntNet in selecting energy-efficient and reliable paths while preserving low computational overhead. The routing decision process considers residual energy, end-to-end delay, packet delivery ratio, and routing overhead, enabling a balanced trade-off between energy efficiency and communication performance. The proposed protocol is evaluated using the NS2.35 simulator under different network densities and traffic conditions. Simulation results demonstrate that the proposed approach reduces energy consumption by up to 32% and routing overhead by 28%, while improving packet delivery ratio by 40% and network lifetime by 22% compared to conventional Ad hoc On-Demand Distance Vector (AODV) and AntNet-based routing protocols. These results confirm that combining AntNet with a lightweight MLP yields an effective and scalable solution for energy-efficient multipath routing in WSNs, without the complexity of deep learning-based schemes.

• View PDF

This study investigates English as a Foreign Language (EFL) students’ perceptions of YouTube's effect as a digital tool on speaking skills, with particular attention to the influence of gender and educational level. The primary objectives are to examine differences in perceptions between male and female students and across varying academic levels within the English Department at Seiyun University. A mixed-methods approach was employed, combining quantitative and qualitative data collection techniques. A questionnaire was administered to 53 students, while semi-structured interviews were conducted with a purposive sample of eight students (four male and four female) representing different educational levels. Descriptive statistics, independent samples t-tests, and one-way ANOVA were utilized to analyze the quantitative data and assess the significance of differences in perception based on gender and academic level. Findings indicate that male students generally reported more favorable perceptions of YouTube’s effect on speaking skills than their female counterparts. Moreover, students at the fourth academic level demonstrated stronger beliefs in the platform’s efficacy in enhancing their speaking proficiency compared to those in lower levels. These results underscore the importance of considering demographic and educational variables when integrating digital media tools like YouTube into language learning curricula. The study recommends that language educators and curriculum designers adopt differentiated strategies to optimize the use of YouTube in speaking skill development, tailoring approaches based on learners’ gender and academic standing.

• View PDF

This study examines the sociolinguistic approach based on Kitābun Marqūm developed in order to improve Arabic digital literacy among university students. Using an explanatory sequential mixed methods design, quantitative data related to learning gains were collected through a quasi-experimental pre-test–post-test control group design, which was then followed up with a qualitative inquiry in order to explain and contextualize the statistical findings. Quantitative results showed that the students who are taught using Kitābun Marqūm's sociolinguistic approach make significantly better gains in digital literacy than those learning through a non-sociolinguistic approach, as shown by t-test and N-Gain scores. Qualitative results demonstrate that the CAF (complexity, accuracy, fluency) gains were due to sociolinguistic awareness, use of digital Arabic texts, and contextualized meaning-making, and indicate limitations related to technology infrastructure and access. Instead of advocating a one-size-fits-all model, the study promotes a context-specific pedagogical approach referred to as Sociotechnical Arabic Literacy (SAL), which encourages sociotechnolinguistic and digital technology literacies and uses AI-supported solutions that aim not only at fostering an informed understanding and critical use of digital media in Arabic but also at promoting a broader sociopolitical awareness of the world. The findings add to the emerging discussion on sociotechnical aspects of language learning in digitally enhanced ecologies.

• View PDF