Cheng-Ji Lai

Language Center, National Chung Hsing University, Taiwan (ROC) // laicj1124@nchu.edu.tw

Vivien Lin

Graduate Institute of Technology and Adolescent English, National Changhua University of Education, Taiwan (ROC) // vivienster@gmail.com

George Martin Berry

Department of Applied English, Chaoyang University of Technology, Taiwan (ROC) // gmberr@gmail.com

Ying-Tung Lin

Graduate Institute of Technology and Adolescent English, National Changhua University of Education, Taiwan (ROC) // m1346007@gm.ncue.edu.tw

ABSTRACT: 

While immersive virtual reality (IVR) and generative AI (GenAI) each offers pedagogical potential, few studies have explored how their integration can support bilingual learners’ construction of scientific explanations in Content and Language Integrated Learning (CLIL) classrooms. This quasi-experiment investigates the effects of a pedagogical GenAI agent in IVR science review games on fifth-grade students’ CLIL science performance, as well as their perceptions on the Task-Technology Fit (TTF). Fifty CLIL learners experienced one of the three conditions: (a) IVR+GenAI review, (b) IVR-only review, and (c) teacher-led review. Data sources included a CLIL science test, oral explanations, surveys, interviews, rater reflections. Findings show that the IVR+GenAI group significantly outperformed the other groups in understanding scientific concepts, particularly for abstract and language-heavy content. For oral explanations, the IVR+GenAI group showed stronger performance in selected dimensions, particularly vocabulary usage, fluency, and communicative confidence in more abstract or language-demanding topics. Rater observations indicated clearer reasoning, more frequent integration of prior knowledge, and more explicit links between scientific concepts and real-world examples in this group. Survey responses further suggested higher perceived task–technology fit among IVR+GenAI students. This study supports the cognitive-linguistic fit in CLIL science IVR games, demonstrates the affordances of multimodal, dialogic GenAI scaffolds in IVR environments, and offers guidance for designing CLIL science IVR games.

Keywords: 

Content and language integrated learning (CLIL), Generative artificial intelligence (GenAI), Immersive virtual reality (IVR) games, Pedagogical agents, Task-technology fit (TTF), Scientific oral explanations

Bilge Has Erdoğan

Republic of Turkey Ministry of National Education, National Education Academy, Başöğretmen Atatürk Education and Practice Center, Türkiye // bilgehas@windowslive.com

Derya Acar Başeğmez

Republic of Turkey Ministry of National Education, Sincan Şehit Abdullah Büyüksoy Art and Science Center, Türkiye // deryaacar03@gmail.com

Yunus Emre Avcu

Balıkesir University, Gifted Education Department, Necatibey Faculty of Education, Türkiye // yunusemre.avcu@balikesir.edu.tr

ABSTRACT: 

In this study, a moderated mediation model was tested in which teachers’ differentiation skills served as the input, innovative practices as the output, grit as the mediator, and technology self-efficacy as the moderator of the relationship between differentiation skills and grit. The research data were collected from 577 teachers working in Türkiye. The latent moderated structural equations approach was employed to test the structural model. The research findings supported the proposed theoretical model. The results indicated that teachers’ instructional differentiation skills both directly and indirectly influenced their innovative practices through grit. Additionally, teachers’ technology self-efficacy was found to moderate the relationship between differentiation skills and grit. The study also revealed a positive relationship between teachers’ education levels and their innovative practices. These results suggest that teacher education should be regarded as a lifelong process and continuously adapted to the evolving demands of the era. By doing so, more innovative practices can be integrated into classrooms to enhance the overall quality of education. In this context, the research findings recommend that institutions and policymakers responsible for teacher education increase the availability of practical programs aimed at strengthening teachers’ pedagogical and technological competencies.

Keywords: 

Teacher innovative practices, Differentiated instruction, Grit, Technology self-efficacy, Teacher training

Adolfo Montalvo-García

Center for Research and Studies for Organizational Development. Universitat Autonoma de Barcelona, Spain // Adolfo.Montalvo@uab.cat

Santiago Ávila Vila

EAE Business School, Spain // Savila@eae.es

David Rodríguez-Gómez

Center for Research and Studies for Organizational Development. Universitat Autonoma de Barcelona, Spain // David.Rodriguez.Gomez@uab.cat

ABSTRACT: 

Teaching presence (TP) is crucial in online education, but its impact in videoconferencing contexts remains underexplored. This study aims to validate a framework that integrates various e-leadership strategies with personalized communication with webcams (PCW). A sample of 592 master’s students in management programmes participated in this research; all of these individuals were subjected to a pedagogical model that consisted of three weekly one-hour videoconferences. This study proposes that TP pertains to two key variables that were examined in this context—PCW and comprehensive e-leadership—on the basis of three different theories (i.e., situational, transformational, and socio-emotional). An exploratory factor analysis identified a valid four-factor model for e-leadership, and structural equation modelling was used to examine the associations between PCW and comprehensive e-leadership. A multigroup confirmatory factor analysis revealed significant group differences in the relationship between PCW-based teaching and situational e-leadership based on students’ learning approach (i.e., content acquisition, collaborative learning, and individual knowledge building through activities). The results also indicate that younger students tend to report higher valuations of comprehensive e-leadership, whereas female students indicate stronger appreciation for socio-emotional and situational e-leadership. The resulting model offers practical guidance for efforts to enhance teaching practices in videoconference-based learning environments.

Keywords: 

E-leadership, Teaching presence, E-communication, Videoconferences

Bo Zhang

Entrepreneurship and Enterprise Hub, XJTLU Entrepreneur College (Taicang), Xi’an Jiaotong Liverpool University, Suzhou, China // Bo.Zhang@xjtlu.edu.cn

Nigel Robb

Research Faculty of Media and Communication, Hokkaido University, Sapporo, Japan // nigelrobb@imc.hokudai.ac.jp

ABSTRACT: 

Children with autism spectrum disorder (ASD) often have difficulties with working memory, attention, inhibitory control, and emotion recognition. These challenges impact daily living and social growth. While computerized cognitive training shows promise, the use of augmented reality (AR) to increase engagement and transfer is not well studied. This randomized controlled trial compared two AR-based interventions—an AR 1-back task and an AR Stroop task—with a traditional 2D Stroop task and a no-training control. Eighty Chinese children with ASD (aged 4–15 years) were randomly placed in one of four groups (n = 20 per group), each completing 10 daily training sessions. Performance was measured before and after the intervention using four CANTAB subtests: Spatial Working Memory (SWM), Delayed Matching to Sample, Multitasking Test, and Emotion Recognition Task. All active training groups showed major within-task improvements and better emotion recognition. Only the two AR-task groups, however, showed notable reductions in SWM between-errors. The 2D Stroop group showed limited transfer, whereas the control group showed no improvement. Between-group analyses showed a significant advantage for the AR-task groups over the no-training control in SWM (ps < .05 after Holm correction). There were no group differences in emotion recognition. These results suggest AR-task-based cognitive training may support broader working memory transfer in children with ASD. However, because no 2D n-back comparison group was included, it's unclear whether the gains stem from the AR format, the task demands, or a combination of both. Replication with fuller designs is needed for definite conclusions.

Keywords: 

Augmented reality, Autism spectrum disorder, Computerized cognitive training, Working memory, Cambridge Neuropsychological Test Automated Battery

Yanyu Yang

Department of Public Courses, Guangdong Police College, Guangzhou 510230, China // 276134682@qq.com // 20130534@gdppla.edu.cn // 

Dan Hu

Department of Public Courses, Guangdong Police College, Guangzhou 510230, China // 13925087181@139.com

ABSTRACT: 

This qualitative study investigates how Chinese university English as a foreign language (EFL) learners strategically manage the visibility of their artificial intelligence (AI) tool use in classrooms. Through scenario-based tasks and interviews with fifteen undergraduates, the analysis revealed three interconnected patterns. First, students employed sophisticated strategies, such as preemptive window minimization and performing visible effort to manage how others saw their AI use. Second, these practices were mediated by significant emotional ambivalence, including anticipatory anxiety, AI guilt, and relief. Third, their technology use was profoundly shaped by cultural scripts valuing visible effort and group harmony. Interpreted through the integrated lenses of facework (Hwang, 1987) and dramaturgical theory (Goffman, 1959), the findings challenge simplistic notions of academic dishonesty by reframing such behaviors as complex cultural adaptations. The study advances theory by proposing three new concepts: pedagogical facework (the strategic performance of face concerns), affective mediation (the emotional process driving such performances), and ethical flexibility (the situated reasoning students use to negotiate AI’s legitimacy). These insights provide a foundation for educators and policymakers to develop more culturally responsive approaches to AI integration in education.

Keywords: 

Artificial intelligence (AI), Facework, Dramaturgical theory, Performative technology use, EFL learners

Hui-Tzu Chang

Center for Institutional Research and Data Analytics, National Yang Ming Chiao Tung University, Taiwan // simple@nycu.edu.tw

Jung-Hong Chuang

Department of Computer Science, National Yang Ming Chiao Tung University, Taiwan // jhchuang@cs.nycu.edu.tw

Chi-Min Hsieh

Institute of Applied Arts, National Yang Ming Chiao Tung University, Taiwan // chimin.hsieh@gmail.com 

Hong-Yu Chang

Department of Communication and Technology, National Yang Ming Chiao Tung University, Taiwan // hongyunycu@nycu.edu.tw

Chuan-Chang Wang

Blacksmith Technology Inc. // kevin.cwang3@gmail.com

ABSTRACT: 

This study compares students in hybrid interdisciplinary practice instruction (HIPI) and in-person interdisciplinary practice instruction (IPIPI) cohorts in terms of classroom engagement, peer assessment, project evaluation, and final grades in the context of an XR cross-domain camp course; it also examines the associations among these variables. A total of 147 students who participated in XR camps from 2019 to 2022 were included in this study (HIPI = 54; IPIPI = 93). These students formed project teams on the basis of their individual interests, such as programming, 3D art, and music, and completed three projects, each of which was followed by a formal presentation on weeks 6, 11, and 16. To support interdisciplinary collaboration, team composition was reshuffled during each stage. Classroom engagement was assessed in terms of attendance, frequency of questions, and the quality of feedback, as recorded by teaching assistants. After each project, the students completed peer assessments and received project evaluation scores, which aimed to encourage reflective practice and team-based learning. The results revealed that students in the HIPI cohort received higher peer assessment and project evaluation scores than those in the IPIPI cohort did, although these differences decreased across project cycles. In addition, for students in both groups, classroom engagement was associated with final grades through peer assessment and project evaluation, and these modeled indirect associations were stronger in the HIPI cohort. These findings suggest that hybrid interdisciplinary practices may provide useful support for assessment and feedback processes in XR-related education. However, because the comparison involved cohorts that were recruited from different years and contexts, the findings of this research should be interpreted with caution.

Keywords: 

Extended reality technologies (XRs), Hybrid interdisciplinary practice instruction, Learning outcomes, XR camp curriculum design

Yi-Fan Wang

Institute of Information and Decision Sciences, National Taipei University of Business, Taiwan // yfwang_tw@ntub.edu.tw

Mei-Hua Hsu

Center for General Education, Chang Gung University of Science and Technology, Taiwan // mhsu@mail.cgust.edu.tw

Max Yue-Feng Wang

Marketing, the Pennsylvania State University, USA // myw5361@psu.edu

ABSTRACT: 

As digital learning environments generate increasingly detailed behavioral and temporal data, machine learning offers new opportunities to model vocabulary acquisition and forgetting processes with greater precision. This study compares three predictive approaches, XGBoost, Long Short-Term Memory (LSTM) networks, and Transformer architectures, for forecasting learners’ post-test vocabulary scores and vocabulary decay over time. Using data from 326 English-as-a-foreign-language learners, the models were trained on study behaviors, test performance, and temporal review patterns. Results show that XGBoost achieved the highest accuracy in predicting post-test scores, reflecting its effectiveness with structured learning analytics data. In contrast, Transformer models outperformed both XGBoost and LSTMs in predicting vocabulary decay, capturing non-linear forgetting trajectories and long-range dependencies in study sequences. Feature importance analyses highlight the central role of prior knowledge, review frequency, and forgetting-curve parameters. These findings demonstrate the value of combining tree-based and sequence-based models to understand cognitive processes in vocabulary learning and inform the design of adaptive, personalized learning systems.

Keywords: 

Vocabulary learning, Machine learning prediction, XGBoost, LSTM, Transformer models

Shu-Hsuan Chang

Department of Electrical and Mechanical Technology, National Changhua University of Education, Taiwan // shc@cc.ncue.edu.tw

Li-Wen Huang

Department of Electrical and Mechanical Technology, National Changhua University of Education, Taiwan // Language Center, Chaoyang University of Technology, Taiwan // phoebecute2000@gmail.com

Xiang-Qin Huang

Department of Finance, National Changhua University of Education, Taiwan // s1167035@gm.ncue.edu.tw

Pei-Ling Chien

Faculty of Engineering, Kyushu University, Japan // chien.pei.ling.019@m.kyushu-u.ac.jp

Po-Jen Kuo

Department of Electrical and Mechanical Technology, National Changhua University of Education, Taiwan // pc7938@icloud.com

Yi-Yun Chung

Department of Electrical and Mechanical Technology, National Changhua University of Education, Taiwan // coopeann@gmail.com

ABSTRACT: 

Generative AI chatbots (Gen-AI chatbots) have become an important emerging technology to promote English language education. Given the mainstream application of Gen-AI chatbots and the accumulation of related experimental studies, a meta-analysis is urgently needed to summarize the current state of knowledge and research gaps in this field. So far, very few meta-analyses have examined the impact of Gen-AI chatbots on English learning achievement, and most existing studies remain limited in scope. This meta-analysis reviews 32 quasi-experimental studies published between 2016 and 2025 to evaluate the effectiveness of Gen-AI chatbot interventions in English language education. The sample contains 115 effect sizes and 3,201 student subjects. Findings reveal a moderate effect size of 0.50, suggesting that Gen-AI chatbots significantly enhance English learning. In addition, this study proposes a “Design framework for applying Gen-AI chatbots in English language education,” which includes moderator analysis with eight variables, including statistically significant variables, such as education stage (ES), school location (SL), language environment (LE), skills targeted (ST), and group interaction (GI). The variables with no significant effect include Chatbot type (CT), Interaction mode (ID), and Treatment duration (TD). The results can be used as a reference for English language education policy makers, educational practitioners, chatbot designers, and academic researchers.

Keywords: 

Meta-analysis, Generative AI chatbots, English language education, Human-computer interface, Technology-based learning

Meva Bayrak Karsli

Department of Computer Education and Instructional Technology, Atatürk University, Turkey // mevabayrak@gmail.com

Muhammed Guler

Faculty of Open and Distance Education, Atatürk University, Turkey // muhammed.guler@atauni.edu.tr

Sinem Cilligol Karabey

Faculty of Open and Distance Education, Atatürk University, Turkey // sinem.karabey@atauni.edu.tr

Melike Aydemir Arslan

Department of Instructional Technologies, Atatürk University, Turkey // melikeaydemir@atauni.edu.tr

ABSTRACT: 

The rapid digitalization of educational environments has substantially increased the volume and diversity of learning-related data, positioning Learning Analytics (LA) as a key and rapidly maturing research field. Despite the growing number of publications, large-scale studies examining the thematic structure and longitudinal evolution of LA research remain limited. Addressing this gap, this study analyzes 7,933 LA-related publications indexed in Scopus between 2010 and 2024 using bibliometric techniques and Latent Dirichlet Allocation (LDA)-based topic modeling. Titles, abstracts, and keywords were processed through natural language processing to identify latent research themes and their temporal dynamics. The findings reveal substantial growth in LA research, particularly after 2015, with more than half of publications produced during the 2020-2024 period. Topic modeling results indicate a balanced and multidimensional thematic structure rather than dominance of a single research focus. Prominent themes include self-regulated learning, student behavioral analytics, predictive modeling, dashboards and feedback systems, artificial intelligence, and ethical considerations. Temporal analyses demonstrate a paradigmatic shift from large-scale online learning environments and MOOCs toward more individualized, multimodal, and affective learning processes. The increasing emphasis on ethical, pedagogical, and institutional dimensions suggests that LA is evolving beyond a purely technical domain into a more reflective, ethically aware, and institutionally integrated research field-exhibiting characteristics of an advancing discipline, though the continued predominance of conference-based publications (57.8%) indicates that theoretical consolidation and methodological standardization are ongoing. This study provides a comprehensive, data-driven overview of LA research evolution, highlighting emerging trends, developmental patterns, and future priorities for researchers, institutions, and policymakers.

Keywords: 

Learning analytics, Topic modeling, Bibliometric analysis, Temporal trends, Educational data mining

Kai-Wen Wu

National Yang Ming Chiao Tung University, Taiwan // Merck Advanced Technologies Ltd., Taoyuan Plant, Taiwan // waskevin0512@gmail.com

Meng-Jun Chen

National Yang Ming Chiao Tung University, Taiwan // vcm1018@nycu.edu.tw

Sheng-Chang Chen

National Yang Ming Chiao Tung University, Taiwan // sengechen@nycu.edu.tw

Ta-Wei Li

Department of Chemistry, National Yang Ming Chiao Tung University, Taiwan // twli@nycu.edu.tw

Hsiao-Ching She

National Yang Ming Chiao Tung University, Taiwan // National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan // hcshe@nycu.edu.tw

ABSTRACT: 

Model-based reasoning is central to scientific and engineering practices for deepening students’ understanding of natural and engineered systems in the 21st century. Yet little research has examined how the cognitive processes underlying model-based reasoning unfold when model comparison scaffolding is provided. With advances in learning analytics and multimodal data analysis, researchers can now capture these processes in fine detail. This study integrates eye-tracking and learning analytics to investigate how model comparison scaffolding shapes students’ visual attention and cognitive focus during digital atomic model–based reasoning. Sixty college students were randomly assigned to either a model comparison scaffolding (MCS) condition or a control condition without scaffolding. Results showed that students in the MCS condition outperformed those in the control group in both multiple-model reasoning and model synthesis. Students in the MCS condition devoted greater attention to advanced models, evidenced by frequent bidirectional transitions between advanced models (Model B ↔ C) and between models and question areas, reflecting more integrative processing. In contrast, control group’s students exhibited mostly linear question-to-question transitions with minimal inter-model transitions. PLS-SEM further revealed that eye-movement behaviors predicted both multiple-model reasoning and model synthesis, ultimately strengthening overall model-based reasoning. Collectively, these findings demonstrate that model comparison scaffolding plays a critical role in directing visual attention and advancing model-based reasoning, while showcasing how multimodal data analysis can illuminate the cognitive mechanisms that support these processes.

Keywords: 

Model-based reasoning, Model comparison scaffolding, Multimodal data analysis, Eye-tracking and learning analytics

Lin Zhao

School of Education, Linyi University, Linyi, China // zhaolinjy@lyu.edu.cn

Xiaohua Deng

Teachers College for Vocational and Technical Education, Guangxi Normal University, Guilin, China // 124658225@qq.com

Ethan (Yi) Cao

Teachers College for Vocational and Technical Education, Guangxi Normal University, Guilin, China // Ethan_tsao@outlook.com

Yanjie Cao

School of Education, Linyi University, Linyi, China // caoyanjie@lyu.edu.cn

Changsheng Ge

Faculty of Education, The National University of Malaysia, Bangi, Malaysia // p125853@siswa.ukm.edu.my

ABSTRACT: 

This study examines how perceived learning support, psychological need satisfaction, digital competence, digital burnout, and academic burnout are related in blended learning environments. Using a mixed-methods design that combined qualitative interviews (N = 16) and a survey of 558 university students in 7 colleges, this study examined the relationships among these constructs within an SDT-informed framework. The results indicate that psychological need satisfaction was positively associated with digital competence and negatively associated with both digital burnout and academic burnout. Perceived learning support was indirectly associated with digital competence and burnout-related outcomes through psychological need satisfaction. Digital burnout was positively associated with academic burnout, suggesting that it represents an important pathway linking students’ digital learning experiences to burnout-related academic outcomes. Overall, the findings support the value of Self-Determination Theory as an integrative framework for understanding how learning support, psychological need satisfaction, and digital learning experiences are associated with student functioning in blended learning. The study contributes to the literature by clarifying the interrelationships among digital competence, digital burnout, and academic burnout, while also highlighting the potential role of psychological need satisfaction within an integrated model of blended learning.

Keywords: 

Distance education and online learning, Improving classroom teaching, Academic burnout, Blended learning, Self-determination theory

Hui-Chuan Chu

Department of Special Education, National University of Tainan, Taiwan // huichu@mail.nutn.edu.tw

Chun-En Yen

Institute of Manufacturing Information and Systems, National Cheng-Kung University, Taiwan // marvinyenn@gmail.com

Pin-Yuan Fang

Institute of Manufacturing Information and Systems, National Cheng-Kung University, Taiwan // j6315617@gmail.com

Wei-Jen Chen

Department of Special Education, National University of Tainan, Taiwan // m72900024@gmail.com

Yuh-Min Chen

Institute of Manufacturing Information and Systems, National Cheng-Kung University, Taiwan // ymchen@mail.ncku.edu.tw

ABSTRACT: 

The advancement of artificial intelligence has introduced real-time and automated capabilities into traditional formative assessment. However, many existing applications remain narrowly focused on academic outcomes and lack the capacity to identify the dynamic and diverse factors influencing the learning process, thereby limiting teachers’ ability to deliver timely, adaptive support in inclusive settings. To address this gap, this study proposes an AI-Driven Progressive Formative Assessment Model with Continuous Optimization. By integrating continuous learning detection with progressive performance assessment, the model diagnoses students’ learning progress, provides real-time adaptive support and preventive guidance, and supports the continuous optimization of learning performance. To operationalize the proposed assessment model, this study developed an attention-based convolutional neural network (CNN) prediction model to predict learning performance and identify salient features associated with the predictions. The prediction model achieved accuracy and F1-score values exceeding 92%. In a separate continuous optimization validation, prediction accuracy increased to 97% as additional learner data were incorporated, indicating the model’s capacity for adaptive updating. A digital reading skills development platform was developed to validate the practical application of the proposed model. The quasi-experimental results showed that students who received instruction through the proposed model significantly outperformed those in the control group, with final exam scores averaging 8.8 points higher (p < .001). These findings provide empirical support for the proposed model’s role in improving digital reading skills and academic achievement among participating students, including those with disabilities, while indicating its potential for adaptation to other academic domains.

Keywords: 

Formative assessment, Digital reading skills, Inclusive education, Continuous optimization, Learning analytics