Yu-Feng Lan

Department of Information Management, National Formosa University, YunLin, Taiwan, R.O.C. // yflan@nfu.edu.tw

ABSTRACT: 

This study investigates how AI-mediated debugging scaffolds can be structurally embedded within automated assessment systems to reorganize novice learners’ problem-solving processes beyond correctness-based feedback. Instructor-authored programming problems were transformed into three adaptive formats: faulty code snippets, line-reordering tasks, and graduated hints. A five-week quasi-experimental study with 90 undergraduates compared a conventional automated assessment environment with an AI-augmented version incorporating debugging-oriented scaffolding. Learners in the AI-augmented condition achieved significantly higher programming performance and more positive attitudes, with particularly pronounced gains among low-achieving students. Log analyses showed differentiated interaction patterns, suggesting that frequent scaffold engagement reflected sustained effort under conceptual difficulty rather than simple performance deficiency. Interview data converged with these results, indicating enhanced confidence, improved error diagnosis, and more reflective reasoning. Conceptually, the findings position AI-generated scaffolding as both cognitive support and an interactional mechanism shaping learner–system engagement in automated assessment environments, supporting process-oriented and learner-centered assessment design.

Keywords: 

AI-generated debugging, Automated assessment, Adaptive scaffolding, Computer science education, Novice learners

Prajakt Pande

Aarhus University, Denmark // prajakt@au.dk

Morten Erik Moeller

University College Copenhagen, Denmark // moem@kp.dk

Biljana Mojsoska

Roskilde University, Denmark // biljana@ruc.dk

ABSTRACT: 

Immersive virtual reality (VR) affords novel sensorimotor experiences that can support embodied learning of complex STEM concepts requiring learners to imagine spatial and dynamic phenomena. However, our understanding of how to harness VR’s unique affordances to systematically implement embodied learning design principles, and how learners’ embodied interactions relate to learning, is still evolving. In this paper, we present the design and development of I, Enzyme, a VR simulation environment conceptualized to foster embodied and enactive learning of complex organic chemistry mechanisms in a biologically grounded enzymatic catalysis narrative. The environment integrates embodied interaction design, phenomenological inquiry, and multimodal learning analytics (interaction- and eye-tracking). We also report findings from a single-group pre–post study involving 41 undergraduates that investigated how learners’ interaction behavior, gaze patterns, and sense of embodiment during simulation use relate to their conceptual understanding of organic reaction mechanisms. We found significant overall pre–post gains in conceptual understanding, intrinsic motivation, and self-efficacy, as well as substantial differences between higher- and lower-gain learners in interaction metrics such as frequency and timing, but not in gaze behavior. The paper contributes to the literature on embodied and immersive learning technology design, and illustrates how multimodal analytics can reveal learning-relevant embodied engagement in VR environments, while also pointing to current methodological constraints of eye-tracking in dynamic immersive settings.

Keywords: 

Virtual reality, Embodied learning, Science education, Multimodal analytics, Eye tracking

Fang-ying Lo

Center for General Education, Asia University, Taiwan // flo@asia.edu.tw

Ya-Tsen Lin

Department of Computer Science & Information Engineering, National Central University, Taiwan // tina30124@gmail.com

Chen-Chung Liu

Department of Computer Science & Information Engineering, National Central University, Taiwan // ccliu@cl.ncu.edu.tw

ABSTRACT: 

Prior empirical studies and systematic reviews have examined interactive reading interventions, including teacher-led dialogic interactions and agent-mediated dialogue, and have reported benefits for learners’ comprehension support and reading engagement through questioning, discussion, and feedback. While tutor agents enabled by generative artificial intelligence (GenAI) have demonstrated efficacy, they often position students in a role to passively answer questions, which may limit learners’ autonomy. This study addressed this gap by exploring tutee agent design, powered by GenAI, to enhance students’ reading interest within the theoretical framework of interest-driven creation (IDC). Within this framework, a tutee agent, acting as a student, aimed to prompt learners to engage in the creation of their own tutee agent by narrative retells, while a tutor agent supported learners’ comprehension. Our study, involving 76 elementary students, compared agent-assisted IDC reading with traditional teacher-assisted and independent reading. Findings supported the positive effects of agent-assisted IDC reading in enhancing students’ interest and flow perception. Notably, interactions with the tutee agent exhibited a positive correlation with reading interest, suggesting that the active process of learning by teaching constitutes a more impactful engagement mechanism than the passive reception of input from tutor agents. Implications of the findings are discussed.

Keywords: 

Interest-driven creation (IDC) theory, Learning by teaching, Agent technology, Generative AI, Reading

Yao-Ting Sung

Department of Educational Psychology and Counseling, National Taiwan Normal University, Taiwan // sungtc@cc.ntnu.edu.tw

Chien-Chih Tseng

Department of Educational Psychology and Counseling, National Taiwan Normal University, Taiwan // s0902041@gmail.com

Tsui-Chun Hu

College of Education, National Taiwan Normal University, Taiwan // tsuichun@ntnu.edu.tw

ABSTRACT: 

With the advancement of adaptive and data-driven technologies, adaptive reading systems have gained increasing attention in educational contexts. However, essential design components for fostering individualized and independent reading remain insufficiently specified. To address this gap, the present study developed and evaluated a Chinese adaptive reading system—the SmartReading (SR) platform—grounded in an integrative framework combining personalized learning and phase-based self-regulated learning (SRL). At the system level, the platform integrates three interrelated components: (a) diagnostic assessment of reading ability, (b) calibrated text recommendation aligned with learners’ proficiency, and (c) structured feedback and reading portfolios designed to support SRL. Using a randomized field trial, 857 fifth-graders and 891 seventh-graders participated in a year–long experiment. The experimental groups used the SR platform weekly alongside their regular Mandarin classes, completing at least six reading tasks, while the control groups continued with regular Mandarin instruction. The results showed that the experimental groups participating in the SR program significantly increased the proportion of students reaching the norm of reading proficiency. Moreover, the reading abilities of students in the experimental group were significantly higher than those in the control group after participating in the SR program for both grades. These findings suggest that adaptive systems grounded in SRL-informed design may function as co-regulatory learning architectures, in which aspects of regulation are shared across learners and adaptive technologies, thereby supporting sustained engagement and reading comprehension in authentic school settings.

Keywords: 

Adaptive reading platform, Independent reading, Self-regulated learning, Text readability, Reading comprehension

Ziyi Wei

University of North Carolina at Chapel Hill, USA // weiziyi@unc.edu

Shihui Feng

University of Hong Kong, Hong Kong // shihuife@hku.hk

ABSTRACT: 

Generative Artificial Intelligence (GAI) tools are increasingly embedded in collaborative learning, yet little is known about how students’ patterns of GAI use relate to their cognitive and metacognitive interactions. Guided by the framework of socially shared regulation of learning (SSRL), this study investigates the associations between specific GAI usage behaviors and cognitive/metacognitive interaction. Seven triads of university students (N = 21) engaged in GAI-supported collaboration tasks. Verbal interactions were automatically coded with a large language model–assisted method, while screen recordings were manually coded for GAI usage behaviors. We used Heterogeneous Interaction Network Analysis (HINA) to examine the associations between GAI use and cognitive/metacognitive interaction, and analyzed post-task focus group interviews to provide complementary insights. The results uncovered how students’ GAI usage behaviors related to cognitive and metacognitive interaction. In particular, GAI usage, in forethought- and performance-phase, including discussing prompting strategies and sharing GAI responses, was strongly linked to concept exploration, planning, and evaluation. Students reported that GAI helped scaffold their thinking, accelerate idea generation, and structure task planning, while also expressing concerns about potential overreliance, which may diminish opportunities for deeper cognitive engagement. This study advances understanding of GAI in collaborative learning by clarifying how usage behaviors connect with cognitive and metacognitive interaction. The findings point to a dual-layered regulatory process in which students regulate both the collaborative learning task and their GAI usage behaviors. It enriches the theoretical understanding of SSRL in GAI-supported collaborative learning and suggests pedagogical strategies for encouraging critical and strategic use of GAI.

Keywords: 

Generative artificial intelligence, Collaborative learning, Cognitive interaction, Metacognitive interaction, Heterogeneous interaction network analysis

Chi-Cheng Chang

Department of Technology Application & Human Resource Development, National Taiwan Normal University, Taiwan // samchang@ntnu.edu.tw

Shih-Chi Chou

Department of Technology Application & Human Resource Development, National Taiwan Normal University, Taiwan // chou.hrd@gmail.com

ABSTRACT: 

This study integrates the revised CoI model and the extended CoI framework to examine the structural relationships among the elements of social media-based autonomy CoI and to confirm its multiple mediation effects. The research samples include 213 adult members in online communities of social media. Research results show that all 6 direct paths and 5 indirect paths in the social media-based A-CoI model reach significant levels and have good model fit. Social presence (SP) and learning presence (LP) not only concurrently and partially mediate the influence of autonomy presence (AP) on cognitive presence (CP), but also serially and partially mediate the influence of autonomous presence on cognitive presence. social presence partially mediates the influence of autonomous presence on learning presence; learning presence partially mediates the influence of social presence on cognitive presence. This study proposes and confirms the multiple mediation model for social media-based A-CoI (AP-SP-LP-CP); i.e., social presence and learning presence have multiple mediation effects between autonomous presence and cognitive presence. These findings have important values and implications for academic and educational practice in CoI research, especially in the environments with weaker instructional functions and unstructured learning. The benefits to researchers and educators are that they can easily monitor, manipulate, and adjust the roles of the four elements in their studies and teaching after knowing the structural relationships among the four elements of online CoI.

Keywords: 

Autonomy community of inquiry, Autonomy presence, Community of inquiry, Multiple mediation, Social media

Hui-Hsun Chiang

College of Nursing, National Defense Medical University, Taipei, Taiwan // huihsunchiang@mail.ndmctsgh.edu.tw

Ching-Long Lai

Division of Basic Medical Sciences, Department of Nursing, Chang Gung University of Science and Technology, Taoyuan, Taiwan // Center for Drug Research and Development, Chang Gung University of Science and Technology, Taoyuan, Taiwan // dinolai@mail.cgust.edu.tw

Wei-Chi Wang

College of Nursing, National Defense Medical University, Taipei, Taiwan // yoyo189467@gmail.com

Gwo-Jen Hwang

Graduate Institute of Educational Information and Measurement, National Taichung University of Education, Taiwan // Graduate Institute of Digital Learning and Education, National Taiwan University of Science and Technology, Taiwan // College of Management, Yuan Ze University, Taoyuan City, Taiwan // gjhwang@mail.ntust.edu.tw

ABSTRACT: 

Generative AI tools like ChatGPT are increasingly integrated into higher education, yet the pathways associated with students’ learning satisfaction remain unclear. This cross-sectional study examined how technology acceptance is associated with learning satisfaction, with learning self-efficacy as a mediator and ChatGPT usage type as a moderator. Participants included 250 undergraduates from five health profession programs at a Taiwanese medical university. Students completed measures of technology acceptance, learning self-efficacy, and learning satisfaction, and answered an open-ended question about ChatGPT’s most beneficial academic use. Thematic analysis identified three usage types: (1) translation and report writing, (2) summarization and integration, and (3) interactive discussion and problem solving. These categories were used as a moderator in a PROCESS-based moderated mediation model. Learning self-efficacy significantly mediated the association between technology acceptance and learning satisfaction, and usage type moderated the self-efficacy–satisfaction association. Specifically, this association was stronger among students who primarily used ChatGPT for translation and writing tasks than among those who primarily used it for interactive tasks. These findings suggest that students’ learning satisfaction in generative AI-assisted learning is not solely associated with acceptance beliefs but also with task-specific usage patterns. While dialogic use may be associated with exploratory learning experiences, practical applications may be more closely linked to confidence and satisfaction. Educators designing AI-integrated curricula should consider the cognitive demands and perceived utility of different forms of AI use to support students’ learning experiences. These instructional design considerations can inform future research linking psychological pathways to objective learning outcomes.

Keywords: 

Generative artificial intelligence (AI), Technology acceptance model, Self-efficacy, Learning satisfaction, ChatGPT usage patterns

Qianru Lyu

College of Integrative Studies, Singapore Management University, Singapore // qrlyu@smu.edu.sg

Wenli Chen

National Institute of Education, Nanyang Technological University, Singapore // wenli.chen@nie.edu.sg

Xinyi Cheng

School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore // xcheng019@e.ntu.edu.sg

Kok Hui John Gerard Heng

School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore // mkhheng@ntu.edu.sg

ABSTRACT: 

Visual joint attention has been identified as a key factor for collaborative learning productivity. However, it is under-studied how students realize joint attention via the intertwined online and offline interactions in the tech-rich collaborative learning environment and how different joint attention levels relate to students’ learning performance. The present study developed and validated an AI-powered computer vision model that integrates deep learning gaze estimation and object detection techniques to support fine-grained analysis of joint attention in face-to-face computer-supported collaborative learning (F2F CSCL). A total of 68 fourth-year engineering undergraduates participated in this study, with their F2F CSCL process video recorded. The evaluation results demonstrated the model’s reliable performance in detecting students’ gaze behaviours across key objects such as laptops, faces, tablets, and worksheets, enabling moment-by-moment identification of dyadic joint attention. Variance analysis reveals that higher performing groups tended to pay more joint attention to screen than lower performing groups while lower performing groups paid more attention to their partner’s screens or partner’s face than higher performing groups. This study enhances the methodological repertoire of multimodal learning analytics by offering a practical and scalable approach to capture joint attention without hardware eye trackers. Furthermore, it advances the theoretical understanding of how joint attention supports knowledge co-construction in technology-rich classrooms.

Keywords: 

AI in education (AIED), Computer-supported collaborative learning, Joint attention, Deep learning, Computer vision

Ekrem Cengiz

Department of Science Education, Bayburt University, Turkey // ec385893@gmail.com

Yasemin Taş Dogan

Department of Science Education, Atatürk University, Turkey // tasyase@gmail.com

Gökhan Aksoy

Department of Science Education, Inonü University, Turkey // aksoygok44@gmail.com

ABSTRACT: 

This study investigated the effect of design-based science instruction on the academic success of 8th-grade middle school students in the simple machine unit, peer learning, metacognitive learning strategies, and attitudes toward engineering. The study was conducted using a quasi-experimental design. The same teacher conducted the study in two randomly selected 8th-grade classes at a public middle school. One of these classes was randomly assigned to the experimental group, which received design-based science instruction, and the other to the control group. LEGOs were used in the experimental group’s design-based science instruction. In the control group, inquiry-based instruction was provided. The study collected data from academic achievement tests, the peer learning scale, the attitudes toward engineering scale, and the metacognitive learning strategies scale. The study was completed in 18 lesson hours. Throughout the study, while the students in the experimental group completed numerous design tasks, the control group received lessons from the same teacher using an inquiry-based approach. The data collected from the study were analyzed using a combination of within-subjects ANOVA. The study found that students in the experimental group had significantly higher scores than those in the control group on academic achievement, attitudes toward engineering, peer learning, and metacognitive learning strategies. As a result of the research, it was suggested that the design-based teaching model be used in teaching other science subjects.

Keywords: 

Design-based science instruction, Learning on simple machines, Peer learning, Metacognitive self-regulation, Attitudes toward engineering

Funda Erdoğdu

Kütahya Dumlupınar University, Türkiye // funda.erdogdu@dpu.edu.tr

Melek Atabay

Trabzon Üniversitesi, Türkiye // melekatabay@trabzon.edu.tr

Ünal Çakıroğlu

Trabzon Üniversitesi, Türkiye // cakiroglu@trabzon.edu.tr

ABSTRACT: 

This study compared pre-service teachers’ perceived learning and assignment-based performance in GenAI-supported and web-based learning environments. A convergent mixed-methods design was used with 70 third- and fourth-year pre-service teachers from various disciplines in an Open and Distance Learning course. Participants were randomly assigned to a ChatGPT group (n = 35) used ChatGPT as their main tool for assignments and a Web group (n = 35) relied on search engines, academic databases, and other online resources. Data were gathered through a perceived learning scale, rubric-based evaluations of assignments, and interviews focusing on students’ assignment experiences and prompting strategies. The findings showed that both groups reported comparable levels of perceived learning. However, students in the ChatGPT group exhibited lower and more inconsistent performance in terms of relevance and accuracy. While perceived learning was strongly correlated with performance in the web-based group, this relationship was weak and not statistically significant in the ChatGPT group, suggesting differences in metacognitive calibration across conditions. ChatGPT group reported iterative prompting behaviors such as rephrasing queries, refining instructions, and conducting sequential checks, whereas the Web group emphasized source evaluation, cross-checking, and synthesis through integrated verification during task completion. Overall, the findings indicated that AI-supported environments may inflate learners’ perceived learning without corresponding gains in actual performance. The findings had implications for teacher education, emphasizing the need to integrate AI literacy, verification skills, and metacognitive scaffolding to foster more accurate self-evaluation and enhance learning outcomes in AI-mediated contexts.

Keywords: 

Perceived learning, Teacher education, Educational technology, Instructional design, AI literacy