This study explores the impact of different VR learning modes (direct vs. vicarious) and instructional sequences (VR-Reading vs. Reading-VR) on middle school students’ science learning outcomes and transmodal moments during VR-integrated station activities. Using a mixed-methods design, the research involved 85 seventh-grade students, including multilingual learners (MLs) and English-first-language learners (EFLLs), who participated in a week-long unit on science lab safety and equipment. A 2 × 2 × 2 quasi-experimental design was employed to examine the effects of VR learning modes or instructional sequences across learner types on science learning outcomes. In parallel, multimodal interaction analysis was conducted on video data from 16 focal participants to examine transmodal moments—instances in which meaning shifted across communicative modes and activities—and their associations with learning outcomes. Quantitative analysis revealed that the direct learning mode resulted in significantly higher knowledge gains compared to the vicarious learning mode. Additionally, there was an indication of a potential interaction between VR learning mode and learner types; however, this observation should be interpreted with caution due to the limited number of MLs and the small effect size. Instructional sequences did not significantly affect learning outcomes. Qualitative analyses showed that the direct learning mode afforded more frequent modal shifts and richer multimodal engagement, while MLs exhibited more transmodal moments in the VR–Reading sequence. However, these transmodal moments did not fully mitigate existing power dynamics between EFLLs and MLs, with EFLLs remaining more dominant in collaborative meaning-making. Overall, the findings suggest that the direct learning mode enhances science learning outcomes by deepening multimodal integration, and that promoting equitable participation for MLs in VR-based collaborative settings may require intentional role assignment and transmodal scaffolding. Implications for designing inclusive, immersive, and multimodal science learning environments are discussed.
Although virtual reality (VR) classroom simulations are recognised as an effective way of promoting teaching skills, they are rarely used for assessment purposes in teacher training. This study examines an analysis strategy rooted in Item Response Tree (IRTree) modelling for estimating the psychometric quality of VR classroom simulations and to identify its implications for designing VR classroom simulations in teacher education, thereby making explicit how the structural properties of a VR classroom simulation determine the requirements of its psychometric evaluation. Data from 200 student teachers who participated in the VR classroom simulation CLASIVIR 1.0 were analysed. The results revealed good model fit and stable estimates, demonstrating a high level of measurement accuracy and strong potential of IRTree models replicating the expected structures of psychological constructs. Thus, the applicability of the proposed analysis strategy was proven, indicating that CLASIVIR 1.0 can be used to reliably measure classroom management competence. Implications derived from the study include suggestions regarding the structure of decision tree-based classroom simulations, sustainability considerations, and issues relating to digital skills.
The rapid growth of online learning has changed higher education, but it has also created new challenges for keeping students engaged. Although digital literacy is often viewed as essential for effective participation, its specific relationship to engagement remains unclear. This study tackles this issue by combining Social Cognitive Theory, the Transactional Model of Stress and Coping, and the Mindfulness-to-Meaning Theory. It examines the indirect pathways between digital literacy affects student engagement in online learning. A multidimensional engagement model comprising behavioural, cognitive, emotional, social, and collaborative components was developed as a higher-order formative construct. Survey data from 300 Malaysian university students in digital learning environments were analysed using PLS-SEM. The results show that digital literacy is positively associated with engagement both directly and indirectly through three mediators, namely self-efficacy, IT mindfulness, and techno-eustress. These findings highlight the importance of what is referred to in this study as “psychological resources”, an interpretive synthesis of the three mediators, in the relationship between digital literacy and engagement. Notably, 78.5
Movement-based learning environments in higher education frequently rely on observational imitation and one-to-many demonstration, which restricts opportunities for individualized feedback in large physical education classes. This study examines the pedagogical impact of PoseSync Vitality, an artificial intelligence based Tai Chi tutor that uses human pose estimation to provide real time movement feedback. The study contributes to research on AI-supported physical education by evaluation of an adaptive instructional system that combined pose analysis, joint specific corrective cues, and practice summaries within the university Tai Chi course. A randomized controlled experiment was conducted with 120 undergraduate students enrolled in a compulsory university physical education course. Students were allocated either to an AI-supported condition using PoseSync Vitality or to a video-based instruction condition following the same curriculum and practice dose. Expert instructors rated performance on a standardized Tai Chi routine at pretest, posttest, and a two-week retention test. Learner experience was assessed with a validated satisfaction questionnaire, and system logs provided a Performance Consistency Score (PCS), an automated index that quantifies the degree of biomechanical alignment between a learner’s movements and an expert reference trajectory. Results show that the PSV group had larger expert rated performance gains from pretest to posttest and from pretest to retention, higher Learner Interface and Personalization scores, and higher posttest PCS values than the video-based group. These results suggest that the PSV instructional system was associated with larger performance gains and higher perceived personalization than video-based instruction in this university Tai Chi context.
Blended learning (BL) is widely regarded as a promising modality for providing flexibility in learning. Yet the psychological mechanisms linking perceived flexibility to students’ cognitive engagement remain poorly understood. Drawing on Control-Value Theory (CVT) of achievement emotions, this study examines how perceived flexibility relates to meaningful and shallow cognitive engagement through control appraisals, value appraisals, and negative-activating achievement emotions (NAEs: anxiety, anger, shame). It also tests whether these associations differ across gender, grade level, and academic major. Survey data from 1,606 Chinese undergraduates were analyzed using partial least squares structural equation modeling (PLS-SEM) with bootstrapping and multigroup analysis (MGA). Perceived flexibility was positively associated with control, intrinsic value, and extrinsic value appraisals. Control appraisal showed no significant association with NAEs. Intrinsic value was negatively associated with NAEs. Extrinsic value was positively associated with NAEs. Anxiety served as the principal mediator. It linked perceived flexibility to shallow engagement through two opposing pathways. MGA revealed that the extrinsic value-to-NAE association was stronger for males. Other paths varied by grade and academic major. These findings refine CVT by identifying boundary conditions for the control-emotion link and by disentangling intrinsic and extrinsic value pathways. Understanding these relationships is crucial for optimizing BL practices and ensuring that they effectively promote students’ cognitive engagement.
This exploratory study examines the potential of artificial intelligence (AI)-based emotion analysis to inform teacher education by analysing preservice teachers’ emotion-related responses during classroom conflict management in virtual reality (VR) simulations. Twenty-two preservice teachers completed three VR scenarios depicting escalating classroom conflicts, generating 66 scenario-based units of analysis. Emotion-related patterns were derived from AI-based analysis of vocal data from the simulations and post-scenario interviews and were considered alongside qualitatively coded coping strategies based on Thomas and Kilmann’s conflict-management framework. The findings provide exploratory indications of dynamic and context-sensitive emotional patterns across scenarios, with Emotional Ambivalence and Emotional Conflict frequently identified at the beginning of the simulations, increased Negative emotions during conflict escalation, and heterogeneous emotion-related profiles after participants implemented coping responses. The comparison between simulation data and post-scenario interviews also revealed discrepancies between AI-derived emotion-related indicators and participants’ retrospective self-reports. Overall, the study highlights the potential of combining VR and AI-based analysis as a multimodal approach to investigating emotionally demanding teaching situations and to supporting research on conflict management in teacher education.
Although a growing body of research has examined students’ attitudes toward generative artificial intelligence (GenAI) in higher education, few studies have compared perceptions across contrasting institutional contexts or explored how students’ reported uses of GenAI relate to broader learning practices. This study addresses that gap by examining university students’ perceptions, self-reported competence, and use of GenAI at two Swedish universities with different academic profiles: a technology-oriented institution and a broader multidisciplinary institution. The study is based on an exploratory questionnaire survey administered to all enrolled students at both universities, yielding 1,097 responses (University A response rate: 11.27
This study compares how students experience and act on feedback when working with an artificial intelligence (AI) peer and a human peer in a project-based assessment. Conducted in upper-level data science courses, the research used a mixed-methods design informed by Feedback Literacy, Evaluative Judgment, and Learning Engagement theories. A total of 103 students completed two equivalent projects, one with a human peer and one with an AI peer, across five stages: ideation, design, coding, debugging, and reflection. Quantitative results showed that AI feedback was rated higher for usefulness, quality, uptake, and trustworthiness, while human feedback supported stronger emotional and social engagement. Qualitative findings revealed that students viewed AI feedback as immediate and practical but sometimes lacking empathy, whereas human feedback provided motivation and mutual understanding. The study concludes that AI and human peers contribute complementary strengths to assessment, with AI enhancing iterative improvement and human peers promoting reflection and collaborative dialogue in learning.
This study investigates the multi-dimensional performance of Notmatix, an AI-powered assessment system, through a mixed-methods design combining quantitative analysis of 764 examination papers with qualitative insights from 67 teachers. The research reveals that AI-assisted assessment is not merely a technical implementation but a complex socio-technical transformation that reconfigures professional identities, institutional power relations, and pedagogical values. Quantitative analysis demonstrated that Notmatix achieved substantial alignment with teacher scoring, evidenced by a strong correlation of 0.80, high reliability with an intraclass correlation coefficient of 0.79, and 90.6
Pre-service teachers (PSTs) often lack adequate opportunities for classroom management practices, particularly in the highly dynamic context of early elementary education. While virtual reality (VR)-based simulation offers a promising alternative, current systems are often limited to semi-automatic interaction, oversimplified generic virtual student design, a text-based modality, and a narrow focus on misbehavior modeling. These limitations compromise the simulations’ capacity to represent classroom complexity and support the natural interactions essential to classroom management practices, thereby reducing PSTs’ sense of presence and engagement. To address these issues, we present SimKids, a novel multi-agent system that utilizes large language model (LLM) and VR, incorporating Big Five Personality Traits and nonverbal behavior design, for classroom management training in early elementary contexts. This study adopts a qualitative, exploratory design to examine pre-service teachers’ experiential engagement with the system. We evaluated SimKids with 27 PSTs in a simulated teaching task and collected their feedback through one-on-one, semi-structured interviews. Findings show that PSTs recognized the virtual students’ authenticity and valued the system’s support for their sense of teaching presence, psychological safety, and reflective engagement with classroom management practices. Their suggestions for integrating socio-cultural elements further revealed a modern conception of classroom management that extends beyond behavior control to address contemporary educational challenges. Our research highlights the perceived affordances of integrating LLMs and VR in multi-agent teaching simulations and provides potential directions for the design and research of classroom management teaching simulation systems.
The article presents an empirical study to test the effectiveness of the Hybrid Interactive Learning with GenAI (HILG) model in improving the quality of programming education in the context of digital transformation of education. The model is based on the multi-agent interaction of five specialized generative AI systems, each performing a distinct pedagogical function: adapting content, providing explanations, diagnosing errors, generating alternative solutions, and facilitating a learning dialogue. The experiment involved 108 students who were randomly assigned to experimental (n = 54) and control (n = 54) groups. The study lasted 15 weeks, and the results showed statistically significant improvements in the experimental group across a number of indicators: average academic performance increased by 25
Generative artificial intelligence is increasingly incorporated into engineering education, particularly in programming courses, raising questions about its effects on learning processes and its acceptability in academic contexts. Empirical evidence examining both interaction patterns and ethical evaluation of Generative artificial intelligence use in real classroom settings remains limited, and studies addressing these dimensions jointly are scarce. This study adopts a multi-phase empirical design to examine student engagement with Generative artificial intelligence across these two complementary dimensions. In Phase 1, a classroom-based activity was conducted with 340 first-year engineering students who completed a time-constrained debugging task using Generative artificial intelligence as the sole external assistance tool. The results indicate that access to Generative artificial intelligence did not ensure reported successful task completion. Instead, successful outcomes were associated with prompt efficiency and verification practices, whereas higher prompting frequency was negatively associated with task success under time constraints. In Phase 2, a qualitative exploratory study was conducted with 16 engineering students through a structured role-play activity simulating an ethics committee, followed by individual voting and survey-based data collection. Acceptance was higher when Generative artificial intelligence was framed as supportive or formative, and lower in scenarios involving summative assessment, surveillance, or autonomous decision making. These preliminary findings suggest that Generative artificial intelligence use in engineering education requires attention not only to tool access or technical performance, but also to students’ interaction strategies, verification practices, and ethical evaluation of context-specific uses. From a socio-technical perspective, the study advances engineering education research by linking technical interaction with GenAI to verification, self-regulation, and ethical responsibility. This highlights the need for pedagogical approaches that integrate technical guidance and ethical reflection into engineering curricula.
With its advanced capabilities, Generative Artificial Intelligence (Gen AI) has increasingly become a tool upon which adolescent students rely. However, whether Gen AI dependency is associated with a heightened impostor phenomenon among adolescent students remains unclear. This study aimed to explore the relationship between Gen AI dependency and the impostor phenomenon among adolescent students while also examining the mediating role of academic self-concept, as well as the moderating role of mindfulness. A total of 2,054 adolescent students (Mage = 18.86, SD = 1.31) completed questionnaires measuring Gen AI dependency, academic self-concept, mindfulness, and the impostor phenomenon. The results indicated that Gen AI dependency was positively associated with the impostor phenomenon among adolescent students, with academic self-concept mediating this relationship. Furthermore, mindfulness moderated the relationship between academic self-concept and the impostor phenomenon. These findings contribute to the field of adolescent digital mental health and offer insights for developing interventions targeting the impostor phenomenon in the AI era.
The pedagogical use of digital technologies in early childhood education is increasingly prevalent today. Augmented reality, virtual reality, and web-based content have the potential to make children’s learning processes more interactive and lasting. In this context, the study compared the effects of different teaching methods presented with augmented reality, virtual reality, web-based, and manipulative (traditional) materials on children’s learning and retention levels. In the quasi-experimental study, a total of 400 children attending preschool education institutions in Bayburt, Turkey, were divided into four different groups and evaluated using a “pre-test–post-test–retention test” model. The results revealed that the virtual reality group achieved the highest scores in both learning and retention levels, followed by the augmented reality and web-based groups, while the manipulative group demonstrated the lowest performance. In addition, the virtual reality environment was evaluated by children as the most enjoyable and satisfying method, and children in this group were the most willing to use the training again. These results show that digital technologies such as virtual and augmented reality in preschool education should be evaluated as effective tools in terms of achieving pedagogical goals and increasing children’s interest in learning.
Artificial intelligence (AI) is increasingly integrated into learning and work, but its association with innovative behaviour remains incompletely understood. Drawing on a sample predominantly composed of higher education students (over 93
This study aimed to design and develop a digital twin-based virtual museum for the Türk Dünyası Application and Research Center (TÜDAM) and, within the framework of the Technology Acceptance Model (TAM), to examine the relationships between information literacy, usability, and satisfaction among its users. Employing a design-based research approach, the study pursued three sequential objectives: (a) designing and developing the virtual museum, (b) assessing users’ perceived usability and satisfaction, and (c) testing the mediating effect of usability on the relationship between IDL and satisfaction. The first objective was addressed by designing and developing a digital twin-based virtual museum, TÜDAM360, using 360-degree panoramic videos and photographs to accurately and immersively represent the physical museum environment; this platform served as the experimental setting for the subsequent phases of the study. To address the second objective, a total of 219 students from Anadolu University experienced the virtual museum using VR headsets and completed scales measuring digital competence, usability, and satisfaction, with data analyzed using K-means clustering and ANOVA. Findings showed that 63.5
This cross-regional study explores preschool teachers’ attitudes toward artificial intelligence (AI) in Türkiye, focusing on their attitudes perceptions, and openness to AI integration in early childhood education. Drawing on data from 405 teachers across all seven geographical regions of Türkiye, the study employed a demographically diverse sample whose gender distribution closely reflects national statistics. Data were collected using a demographic questionnaire and the General Attitude Scale Toward Artificial Intelligence. In this study, the relationships between demographic variables and behavioral factors (AI usage and AI-related training) were examined using correlation analysis, and the findings were interpreted within a relational rather than a causal framework. Findings indicate that preschool teachers generally hold positive attitudes toward AI, particularly recognizing its educational benefits and potential to improve daily life. However, some reservations persist regarding ethical concerns and the preference for human interaction in early learning settings. Statistical analyses revealed significant differences in attitudes based on factors such as education level, region, professional experience, type of institution, digital technology competence, daily internet use, and taking educational technology courses. Younger teachers and those with higher digital competence or frequent use of AI tools demonstrated more favorable perceptions. It is understood that examining only demographic factors is not sufficient; when behavioral factors (such as internet and technology use and participation in training) are also analyzed and evaluated in relation to demographic variables, the factors influencing AI attitudes can be identified more reliably. This, in turn, strengthens the validity and robustness of the study’s findings. These results suggest that while early childhood educators are open to adopting AI technologies, targeted professional development and equitable access to technological resources are essential to support effective implementation. Implications for policy, teacher training, and future research are discussed.
Cognitive overload during technology adoption represents a significant challenge in STEM/STEAM education, potentially affecting teachers’ acceptance and effective use of educational technologies. To better understand this issue, our study aimed to adapt and validate the TAM-CLT model for teacher purposes and examine how cognitive load influences technology acceptance in STEM/STEAM learning environments. Using Rasch measurement and Structural Equation Modeling (SEM), data from 801 STEM/STEAM teachers were analyzed, including an evaluation of rating scale category functioning to ensure the accurate interpretation of response criteria. The results indicate that the model demonstrates strong psychometric structure, performs reliably, and effectively captures meaningful variation in respondents’ perceptions. Furthermore, gender, teaching subject, and technology type were found to moderate several relationships among TAM and CLT constructs. These findings provide a methodological framework for examining educational stakeholders’ perceptions of technology adoption and cognitive load. The study also highlights implications for educational practice, further validation across diverse contexts, and future extensions of the model. Clinical trial registration. Not applicable.
Students’ uptake of peer feedback enhances their learning, but we do not know the antecedent behaviors that shape uptake (apply, copy, reject, or ignore). Hence, we analyzed the 29,545 behaviors of 135 university sophomores during online interactive peer assessment (OIPA) of three activities. Multivariate outcome, multilevel cross-classification analysis showed that external factors (e.g., demographics), synchronous (vs. asynchronous) discussion mode, author agreement or disagreement, author or peer explanations, and specific feedback boosted author applications of peer feedback. After receiving specific feedback, authors copied more peer feedback. Disagreements, requests for specific information, or idea exchanges in the last phase of author-peer discussions yielded more rejections of peer feedback. After a peer ignored the author’s specific feedback requests at the beginning of OIPA, the author ignored more peer feedback. These findings show how dialogic feedback behaviors and specific feedback during OIPA shape students’ uptake of peer feedback.
This study examined the digital literacy (DL) profiles of pre-service social studies teachers in Turkey and investigated how background characteristics predict profile membership. Data were collected from 534 participants across six universities using the Digital Literacy Scale. Applying factor mixture modeling, a three-class structure was identified: high (15.9
