ABSTRACT. Generative AI can provide learners with explanations and answers easily, but such output may be accepted without sufficient meaning construction. This study proposes necessitation of meaning construction as a design principle for learning with AI-generated output. Based on this idea, we designed a two-stage task in which learners first construct a concept map from an AI-generated explanatory text and then reconstruct it from decomposed components. The task is intended to prompt learners to externalize and reexamine conceptual relationships in AI-generated explanatory texts. To explore this design, we conducted an exploratory study to examine question generation across phases, revisions of conceptual relationships in maps, and learners’ subjective evaluations. The results showed that the construction and reconstruction phases were perceived to contribute more to deeper understanding than the initial reading phase, and that both learners’ questions and concept maps changed across phases. These findings provide initial evidence that the proposed task may promote active reinterpretation of AI-generated explanations.

ABSTRACT. LLM tutors are usually optimized for correctness, but how correct knowledge is disclosed may also shape learning. We decompose Academically Productive Talk (APT) into three prompt-conditioned dimensions: Learning Community (LC), Knowledge Precision (KP), and Rigorous Thinking (RT). We implement these dimensions in an LLM tutor for Signal Detection Theory. In a pilot component-ablation study (N = 42), university students learned SDT and then interacted with one of four tutoring profiles. Low-LC produced lower immediate memory and higher perceived workload, while Low-RT produced lower memory with little short-term difference in transfer. Low-KP preserved factual accuracy while reducing immediate use of canonical formulations. It produced higher perceived workload; motivation and memory moved in opposite directions, although neither contrast reached conventional significance. A post-hoc audit using 18 fixed Chinese learner probes documented distinct outputs across prompt configurations. Together, these findings motivate treating pedagogical dialogue as a set of adjustable conversational levers.

ABSTRACT. Although the use of artificial intelligence (AI)-generated feedback (FB) is promising for enhancing learning, evidence showing how its internal composition affects student learning remains scarce, in particular in elementary education. Most previous studies treat FB as a categorical variable instead of decomposing its characteristics as continuous predictors. Longitudinal lag effects—the ways through which FB in one lesson influences reflection in the next—remain underexplored. This study investigates how AI-generated FB composition, which is operationalized through Hattie and Timperley’s FB model [including the levels of Process (FP), Self-regulation (FSR), and Self (FS)], longitudinally affects the quality [Structure of the Observed Learning Outcome (SOLO) scores] and quantity (character count) of social studies reflections by elementary students. We performed a 6-month design-based research field study (August 2025–February 2026) in a Japanese public elementary school (N = 14 students, 691 observations, 64 class sessions). The FB system, implemented using GPT-4.1 mini and GPT-5.1, was iteratively refined through three phases of the design (Types 1–3). The numbers of FP, FSR, and FS aspects per FB text were coded as continuous variables and entered into linear mixed models as lagged predictors. First, FP aspects in previous FB exhibited consistent positive correlations showing subsequent reflection quality and quantity (ρ = .102–.168). Second, FSR significantly predicted character count (β = 5.44, p = .009), but FS negatively predicted SOLO scores (β = −0.162, p = .022; marginal after false discovery rate correction, q = 0.066). Third, growth trajectories accelerated in Phase 2, when FP and FSR increased and FS decreased in prompt redesign (Phase 2 × class session: β = 0.051, p < .001 for SOLO); this elevated level was sustained through Phase 3. Because FB Type and time period are confounded and Phase 3 incorporated a model upgrade, strict causal attribution to FB redesign requires additional controlled studies. The decomposition of AI FB into continuous aspect-level variables produces differential dose–response relationships across FB levels, as well as supporting design guidelines that minimize Self-level praise and enrich Process and Self-Regulation content.

ABSTRACT. In recent years, Virtual Learning Environments (VLEs) have generated large volumes of educational data, enabling the development of intelligent systems for predicting student performance and improving learning outcomes. Early prediction of academic performance is particularly important for identifying at-risk learners and enabling timely pedagogical interventions. However, existing approaches often suffer from a lack of interpretability, limiting their practical applicability in real educational settings. Moreover, most predictive models fail to provide actionable recommendations that can effectively support both instructors and learners. To address these challenges, this paper proposes an interpretable hybrid framework for early multi-class academic performance prediction in virtual learning environments. The proposed approach integrates deep learning (TabNet) and ensemble learning (LightGBM) through a feature fusion strategy that combines original input features with learned latent representations. This hybrid architecture enables the model to capture both low-level and high-level patterns in educational data, thereby improving predictive performance and robustness. In addition, we introduce an enhanced explainability method, termed Adaptive Sharpened SHAP (AS-SHAP), which improves the clarity and concentration of feature contributions by applying a non-linear transformation. The resulting explanations are further transformed into an interpretable embedding space that characterizes each student's learning profile. Based on this representation, a fuzzy logic-based recommendation system is developed to generate personalized and actionable pedagogical feedback using linguistic rules. The proposed framework is evaluated on a real-world educational dataset under a progressive early prediction setting, simulating different stages of the learning process. Experimental results demonstrate that the proposed approach achieves high predictive accuracy while significantly improving interpretability and decision support capabilities compared to conventional methods. Overall, this work provides a unified and interpretable framework that bridges prediction, explanation, and recommendation, contributing to the development of intelligent and trustworthy e-learning systems.

ABSTRACT. This exploratory study re-analyzed student-chatbot programming conversations to examine whether observable interaction patterns differed across learning-gain groups. The dataset comprised 50 students, 1,550 message turns, and 96 conversation sessions. Students were classified according to the direction of change from pre-test to post-test scores: increase (post-test score higher than pre-test; n = 27), same (no score change; n = 14), and decrease (post-test score lower than pre-test; n = 9). Using rule-based behavioral coding, sequence-oriented uptake analysis, student-level feature aggregation, and qualitative case comparison, the study examined both broad interaction patterns and specific forms of student follow-up after chatbot support. Broad quantitative comparisons showed descriptive but mostly non-significant differences across groups. The increase group showed the highest mean rate of code revision after assistant support (0.053), while the decrease group showed the highest mean rate of verification after support (0.131), with verification after support approaching but not reaching conventional significance (p = .057). Refined uptake analysis further suggested that broad productive follow-up was too general to separate groups clearly: increase and decrease students showed similar transformative uptake, while the decrease group showed the highest checking/help-seeking uptake. Three qualitative cases illustrated revision-oriented, checking-oriented, and mixed or fragmented interaction trajectories. These findings suggest that learning-gain groups were not clearly distinguished by overall chatbot use alone; rather, the form of student uptake after chatbot support may be a more informative descriptive lens. The study is framed as exploratory and does not make causal claims.

ABSTRACT. Input device proficiency may act as a hidden barrier in data literacy education, where students struggle with pointer operations rather than statistical content. A prior study (N = 11) found that mouse operation time was significantly correlated with Excel assessment scores (r = −.692, p = .018). The present study compared mouse and touchpad use by analyzing EEG-based cognitive load and pointer trajectory data from the same experiment (N = 10). Theta/alpha ratio was used as a cognitive load index. Pause count and pause time were extracted as hesitation indicators. Touchpad operations produced significantly more pauses than mouse operations across all six tasks (Holm-corrected p = .030, Cliff's d = 0.64–1.00), with the largest differences observed in CopyPaste and Drag tasks. Pause count was correlated with Excel scores in both conditions, suggesting hesitation affects scores regardless of device. However, EEG theta/alpha ratio was associated with Excel scores only in the touchpad condition (rs = −.690, p = .027), not in the mouse condition. These findings suggest that lack of input device proficiency may impose cognitive demand that affects learning outcomes. Implications for instructional design are discussed.

ABSTRACT. Handwriting logs, including pen pressure, offer less intrusive data because they can be collected during writing activities, without requiring separate monitoring devices. Although pen pressure has been linked to engagement-related states, most studies have examined it in controlled settings, leaving its relevance in naturalistic classroom contexts largely underexplored. Yet, real classroom situations may shape learning engagement differently, especially for students with special needs, whose participation may depend on their individual characteristics and classroom support. Therefore, it is necessary to examine whether pen pressure can provide meaningful insight into students’ engagement during regular learning activities. This study explores pen pressure data from six special needs students, collected on a digital learning platform over a one-month period of music therapy lessons. Two aspects were investigated: how pen pressure differs across student attention levels, and how this information was interpreted by the teacher in relation to students’ characteristics. Pen pressure intensity, variability, and within-task change patterns were analyzed across students in high and low attention groups based on teacher ratings. Results showed that pressure variability distinguished attention groups more clearly than pressure intensity. Fluctuating patterns were observed exclusively among low attention students, particularly during tasks with less teacher guidance. Reflexive thematic analysis of the teacher interview suggested that pen pressure patterns may reflect students’ affective states, cognitive effort, and transitions in attention. Overall, when combined with teacher interpretation, pen pressure data may serve as behavioral indicators of students’ learning engagement during handwriting-based tasks. These findings suggest their potential as a foundation for developing automated learning monitoring systems in special needs education.

ABSTRACT. In intelligent tutoring systems (ITSs), learners may enter learning deadlocks, often referred to as wheel-spinning, where productive learning stalls despite continued interaction. While prior work has explored affective and cognitive indicators of such states, feedback triggering is still commonly based on observable performance outcomes. This paper proposes a sensor-agnostic architecture for in-process feedback that explicitly separates when to intervene from what feedback to provide. Intervention timing is inferred from process-level signals such as affective and embodied cues, while feedback content is conservatively diagnosed using an explicit symbolic domain model. Building on prior work using EEG-based timing inference, we examine whether the same architectural principle holds with vision-based sensing. We implement the proposed architecture in an ITSs for arithmetic word problem posing, using a Transformer-based vision model to detect potential learning deadlocks and a domain knowledge model to diagnose content-level issues. A small-scale preliminary evaluation with university students indicates that, although timing detection is imperfect and exhibits a high false-negative rate, the feedback content generated when intervention occurs remains appropriate and non-disruptive. These results suggest that separating timing inference from content diagnosis provides a robust and safe design principle for in-process feedback, enabling practical deployment across different sensing modalities.

ABSTRACT. Generative AI can readily provide working code, creating a tension in introductory programming: students need timely support, yet direct solutions may reduce opportunities for reasoning. To address this, we propose example-based scaffolding, in which GenAI generates scaffold examples that share a target task’s underlying reasoning pattern but differ in surface context, supporting analogical transfer while reducing the risk of direct copying. Specifically, we contribute a two-dimensional taxonomy of scaffold examples with operationalized criteria for identifying effective scaffolds, and CodeExemplar, a prototype that integrates scaffold example generation into a web-based coding environment with automated test-case grading. We report findings from a formative evaluation consisting of a classroom pilot with 30 students and a prototype evaluation with 13 students and 4 instructors. Students perceived scaffolded examples as useful for making progress without revealing the target solutions and preferred them over other forms of support. Instructors also viewed the approach as promising for transfer-focused practice. We discuss design implications, limitations, and directions for future studies.

ABSTRACT. Large Language Models (LLMs) are increasingly used to generate educational content, but their effectiveness as pedagogical designers remains unclear. We conducted a blinded, randomized study comparing three frontier models on probability question generation across five prompting strategies: Basic Difficulty, Bloom’s Taxonomy, Webb’s Depth of Knowledge (DOK), Graesser’s Deep Reasoning, and Contextual Transfer. Twenty Teaching Assistants evaluated outputs for mathematical integrity, instruction adherence, and pedagogical value. The results revealed a trade-off between reliability and educational utility. GPT-5.1 was the most reliable, achieving 100% mathematical integrity but producing formulaic, lower-engagement content. Grok 4.1 received the highest pedagogical ratings but had a 14.7% mathematical error rate. Gemini 3 Pro generated more diverse questions but followed instructions less consistently. Simple difficulty prompts failed to meaningfully differentiate outputs, while structured taxonomies provided better control. Overall, no single model is sufficient for automated question generation (AQG); educators must balance reliability and pedagogical engagement according to instructional goals.

ABSTRACT. Lesson plan alignment, the consistency of cognitive depth between teaching objectives and instructional procedures, is critical for developing higher-order thinking. However, existing lesson plans often suffer from cognitive misalignment, wherein high-standard objectives are reduced to low-level tasks in practice. This study employs computational text analysis to compare cognitive alignment between human-authored lesson plans (HLPs) and Multi-Agent System (MAS) generated plans. Using a corpus of 288 instructional segments, targeted dictionary matching mapped instructional verbs to Webb’s Depth of Knowledge (DOK) framework, while Structural Topic Modeling (STM) identified macro-level pedagogical distributions. The results reveal two key findings. First, at the micro-lexical level, HLPs exhibit significant cognitive misalignment, heavily relying on DOK Level 1-2 verbs despite setting higher-order objectives. Conversely, MAS-generated plans demonstrate a commanding presence of Level 3-4 analytical verbs. Second, at the macro-structural level, STM analysis uncovers a critical pattern of content convergence alongside pedagogical divergence. While HLPs and MAS align perfectly in factual content coverage, MAS significantly outperforms HLPs in constructing explicit higher-order instructional scaffolds, effectively avoiding the pervasive “activity trap.” Ultimately, this study redefines cognitive misalignment as a procedural translation gap and proposes a data-grounded workflow for human-AI collaborative instructional design. It provides empirical evidence that while human educators are essential for contextual anchoring, the MAS functions as a vital structural enforcer, empowering teachers to design rigorous learning experiences without weakening pedagogical alignment.

ABSTRACT. Peer assessment has been widely recognized as an effective pedagogical approach in higher education, fostering critical thinking and evaluative competencies. Central to its effectiveness is the quality of feedback that learners provide and receive. However, feedback engagement is shaped by learners' individual psychological and cognitive characteristics. Among these, self-efficacy and cognitive load have been theorized to influence how learners interact with, process, and act upon feedback. Despite growing interest in feedback literacy, the mechanisms through which these factors operate within interactive peer assessment contexts remain insufficiently understood. To address this gap, the present study examined the effects of self-efficacy and cognitive load on feedback quality and feedback literacy in a dual-cycle peer assessment context. Twenty-four undergraduate students participated in three rounds of peer assessment, generating 198 assessment records alongside pre- and post-course questionnaires. Epistemic Network Analysis (ENA) was employed to examine cognitive network structures underlying feedback quality, while one-way ANOVA and LSD post hoc comparisons were used to assess group differences in feedback literacy dimensions. The findings revealed that learners with higher self-efficacy significantly outperformed their peers in Eliciting, Processing, and Enacting, and demonstrated more complex cognitive networks shifting from personal opinion toward capability assessment and supportive construction. Regarding cognitive load, higher levels were significantly associated with lower scores in Processing and Readiness to Engage, while a moderate level of cognitive load fostered the most balanced and fully developed connections among cognitive elements in feedback interactions. These findings offer practical reference value for the design of peer assessment activities and the cultivation of feedback literacy in higher education settings.

ABSTRACT. We introduce a web-based simulation for teacher professional development that makes LLM training processes transparent through explicit controls and implicit preference‑based shaping. Designed with teachers’ formative feedback practice as the context, our simulation positions teachers as AI co‑trainers. A qualitative pilot with three experienced in‑service teachers showed building up of clearer, more process‑focused intuitions about how LLMs learn. Interaction traces also revealed teachers' increased noticing of how tuning and critiques shape outputs.

ABSTRACT. Generative AI is increasingly positioned as a tool for personalised learning, yet less is known about how young learners participate in shaping AI-mediated support in classroom practice. Building on prior conceptual work on agentic personalisation (AP), this study examines how AP was enacted and constrained in Korean elementary science classrooms. AP is defined as a learner-authored and teacher-mediated process in which students shape learning goals, orchestrate AI-supported pathways, evaluate AI-generated responses, and transform AI-generated outputs into their own learning artefacts. Drawing on an ethnographically informed qualitative case study, the main analysis focused on 41 Grade 6 students who used GenAI during lessons on plant structure and function. Data included student worksheets, AI dialogue records, reflections, final written explanations, and teacher-generated observation logs. Eight focal cases were selected through maximum variation sampling to capture contrasting patterns of AI engagement. Findings show that some students enacted AP beyond prompting by questioning the factual reliability of AI responses or transforming AI-generated explanations into meaningful analogies. However, productive prompting did not always lead to output ownership; some students relied on AI wording, repeated vague questions, or drifted off-task. The study contributes to AI in education research by showing that AP is a learner-authored, teacher-mediated, and unevenly enacted process rather than a form of system-delivered adaptation.

ABSTRACT. Agentic artificial intelligence (AI) systems are autonomous, goal-directed tools capable of decomposing tasks, generating intermediate reasoning steps, and dynamically adapting within learning environments. These capabilities are increasingly shaping how learners engage in problem solving and knowledge construction. While such systems can improve task completion and provide structured guidance, their influence on learning processes remains insufficiently understood. In particular, limited attention has been given to how overreliance on agentic AI systems may allow learners to bypass knowledge construction processes, such as problem analysis, reasoning, and strategy development that typically occur during learning, and how this may influence learning outcomes, including retention and knowledge transfer. This study develops a conceptual framework to examine whether agentic AI functions as a cognitive amplifier that supports comprehension, or alternatively contributes to a comprehension illusion in which task performance improves without gains in durable learning outcomes. The framework integrates two complementary theoretical perspectives: Cognitive Load Theory and Self-Regulated Learning Theory. From a cognitive load perspective, AI-generated reasoning steps may reduce extraneous cognitive load by simplifying problem-solving processes. However, overreliance on such support may reduce germane cognitive effort required for schema construction, retention, and knowledge transfer. From a self-regulated learning perspective, AI-generated solutions may reduce learners’ engagement in planning, monitoring, and evaluation. The design will examine differences in cognitive load, self-regulated learning processes, and learning outcomes, including retention and transfer.

ABSTRACT. This study proposes TriFusion-EduVec, a multimodal transformer model that jointly represents statistical features, behavior sequences, and time-interval information to improve both the performance and the post-hoc interpretability of dropout prediction. Instead of treating timestamps superficially, we explicitly model the temporal rhythm of student engagements. Empirical evaluation on two large-scale MOOC datasets demonstrates that TriFusion-EduVec significantly outperforms recent baselines. More importantly, through an in-depth analysis of the model's attention distribution and masking experiments, we find that the critical indicator of student dropout lies not in the execution of individual actions or specific behavior sequences, but in the underlying learning rhythms and time-interval dynamics. While dropout students exhibit fragmented, high-frequency, and jittery rhythms, persisting students maintain a structured and stable learning cadence. These empirical findings provide solid statistical interpretations that align with self-regulated learning and cognitive load theories, offering researchers and platform designers deep insights into how temporal regularities, rather than behavioral content alone, reflect learning persistence.

ABSTRACT. Classroom video offers rich insights into teaching and learning processes but remains underutilized due to challenges in scalability, accessibility, and data preprocessing. This paper presents COREference, an AI-enabled approach for transforming large-scale classroom video data into a curated corpus for research and teacher professional learning. The study addresses a critical bottleneck in classroom video corpus construction: the reliability of audio quality and transcript generation. We propose a two-layer architecture that separates automated data curation from human–AI interaction. The first layer implements an AI-supported preprocessing pipeline that integrates multi-feature acoustic analysis (RMS energy, spectral centroid, and spectral flatness) for audio quality assessment, alongside domain-specific automatic speech recognition (ASR) model selection for transcript generation. The second layer implements a large language model-based framework (Qwen3-4B-Thinking-2507) to support semantic retrieval and prompt-guided pedagogical code inference using aligned video frames, transcripts, and expert-defined pedagogical code schemes. Current findings show that audio quality is a key determinant of transcription accuracy, with multi-feature filtering effectively improving downstream ASR performance. Comparative evaluation of ASR models across subject domains further reveals that transcription quality varies by language and classroom conditions, motivating subject-specific model selection. Preliminary implementation of the Layer 2 framework demonstrates the feasibility of using LLMs to support structured pedagogical coding and semantic exploration of classroom data, while reducing reliance on manual annotation processes. The proposed approach supports improved corpus reliability and usability, highlighting the potential of AI-supported preprocessing for scalable educational research infrastructures.

ABSTRACT. Blended learning has become a prominent modality in higher education due to its high degree of spatiotemporal flexibility and autonomy. However, spatial transitions and highly autonomous environments pose challenges in maintaining high levels of student engagement, and traditional static evaluation metrics struggle to capture the dynamic evolutionary patterns of engagement during the learning process. Therefore, this study innovatively introduces Dynamic Systems Theory (DST) to conduct an intensive longitudinal tracking of the multidimensional engagement trajectories of 30 undergraduate students over an 8-week (16 observations) blended learning course. Utilizing analytical methods such as moving extreme value plots and moving correlation coefficient plots, the study reveals that: (1) student engagement in blended learning environments exhibits nonlinear development. Specifically, dimensions like cognitive engagement align with the U-shaped development pattern in DST, and external environmental transitions are prone to trigger unstable systemic oscillations. (2) Complex self-organizing relationships exist among the behavioral, emotional, and cognitive sub-dimensions, manifesting in three typical development models: "behavioral-supported", "emotional-competitive", and "fully-supported". This study provides a novel dynamic perspective for understanding student learning mechanisms in blended learning and offers empirical support for implementing targeted pedagogical interventions.

ABSTRACT. Concept mapping is a knowledge-building activity that requires learners to interpret concepts, articulate relationships, and integrate new information, yet little is known about how the process of building a map relates to cognitive engagement and outcomes. In this study, we examined how behavioral engagement relates to perceived cognitive load and learning performance, and used transition network analysis to characterize action dynamics. Results show no straightforward linear relationships among behavioral engagement, cognitive load, and performance. Action sequence analysis further revealed that page flipping and object moving dominated students’ behaviors, potentially overshadowing more meaningful action patterns. These findings highlight the need to elaborate the pedagogical framework to model student behaviors and develop support or interventions to encourage productive engagement during concept mapping.

ABSTRACT. The assumption that generations born in the twenty-first century naturally possess digital literacies remains contested. This study is driven by the continuing disagreement observed in empirical studies on whether students need to be digitally instructed. Via the lens of the Dunning-Kruger effect, the current study explores the disparities between self-perceived and tested digital literacies among English major students in the Chinese higher education sector. 227 students participated in one self-reported questionnaire and an actual test regarding digital competence. Grouping by quartile in test results, discrepancies were observed by comparing the standardised values of self-reported digital literacies with the scores in the tested digital literacies. The ANOVA analysis indicates a significant Dunning-Kruger effect (p < .001, η² = .416), with the low-score group (Quartile 1) overestimating their digital competence (M = 1.17, SD = 1.15), and the high-score group (Quartile 4) underestimating it (M = -1.24, SD = 0.92). The Tukey post-hoc analysis supports significant differences between all groups in pairs. It is suggested that systematic misperception can be found in EFL students’ digital literacies, challenging the meaning of digital natives. This implies that digital literacies education needs to pay attention to self-recalibration in students.

ABSTRACT. Japan's growing multilingual school population has intensified the need for effective communication between schools and foreign guardians. While language barriers are widely recognized, a more fundamental challenge lies in the implicit sociocultural knowledge embedded in Japanese school communication—knowledge that conventional translation tools fail to convey. To address this gap, this study developed PAPAMAMA-TOMO, an AI-mediated multilingual support system designed to help foreign guardians interpret Japanese school culture through contextually grounded explanations. The system integrates a school-print knowledge base constructed from 2,754 school handouts, from which 1,792 implicit knowledge entries were extracted using NLP and LLM-assisted methods, combined with a Retrieval-Augmented Generation (RAG)-based conversational interface supporting multiple languages and image-based input. A preliminary pilot study was conducted with 26 foreign guardians, primarily Vietnamese-speaking (50%) and Nepali-speaking (42%), the majority of whom had lived in Japan for more than three years. Among the participants, 81% reported having used the system. Results indicated that approximately 70% of users experienced a reduced need to contact schools for clarification, and approximately 77% reported increased confidence in handling school-related communication. The explanation function was most valued, followed by translation support and situational guidance. These findings suggest that AI-supported sociocultural interpretation—beyond translation alone—holds significant promise for promoting educational inclusion in multicultural school communities.

ABSTRACT. Pre-writing is the initial step in writing a persuasive English paragraph. However, this task proves challenging for beginners of English paragraph writing. Therefore, in the previous research, a support prototype system for organizing an English paragraph outline was developed and utilized in actual classrooms. During the classroom practice, some students were found to be stuck in the idea organization phase. To address this problem, we intend to implement a new feature called Intent-Adaptive Support into the support system. Realizing this feature requires three functions, which are "operational state recognition," "intent inference," and "adaptive support." As a first step toward implementing Intent-Adaptive Support, this research aims to develop the function of operational state recognition. This function is designed to track the user's progress. Specifically, it recognizes the user’s operational state by monitoring the user’s interactions. Our evaluation showed that applying operational state recognition to user log data enables the system to understand the user’s situation.

ABSTRACT. This study evaluates the pedagogical viability of LLM-generated content within a localised English as a Foreign Language (EFL) context. Utilising log data from Japanese junior high school students practicing on a grammar drilling application, we analysed how different question modalities impact student performance and whether theoretical localised CEFR difficulty tiers accurately predict empirical task difficulty. Results reveal a clear performance hierarchy: multiple-choice questions carried the lowest cognitive load, cloze tasks posed the greatest barrier to active recall, and drag-and-drop exercises incurred the heaviest time penalties. Furthermore, learner data validated the CEFR-J grammar framework, showing a steady decline in accuracy and increased response times as proficiency levels advanced. These findings demonstrate that LLMs can successfully generate learning content, while highlighting the need for developers to strategically sequence question modalities to transition learners from passive recognition to active linguistic production.

ABSTRACT. Collaborative learning remains valuable in the AI era as the social-interactional aspects of learning are still irreplaceable, while its classroom implementation remains challenging due to the burden group activities can place on teachers. Though data-driven technologies have increasingly supported individual learners through explainable intelligent systems, interpretable and actionable AI support for group learning remains underexplored, especially in K12 classrooms. To bridge this gap, this study proposes an explainable data-driven group learning paradigm where LLM-based narratives complement algorithmic group formation and group awareness via Wordcloud. A pilot study was conducted with second-year Japanese senior high school students from basic and advanced EFL classes using a crossover design. The results showed an inconsistent objective advantage of the narratives and might suggest additional cognitive load for higher-performing learners. Lower-level learners perceived the support as more helpful, indicating the feasibility and conditional value of translating learner analytics into human-readable support for classroom group learning.

ABSTRACT. While peer learning is a highly effective collaborative educational approach, digital platforms often struggle with low participation and unresolved questions. To address these challenges, this paper presents a novel approach within the field of Artificial Intelligence in Education (AIED) by integrating Large Language Model (LLM)-based simulated peers into an online peer learning system. Unlike traditional pedagogical agents that act as authoritative instructors, these data-driven simulated peers function as reciprocal learning companions to provide adaptive learning support within a human-AI co-existing community. Grounded in Open Knowledge and Learner Model (OKLM) which tracks individual learner proficiency from learning logs, the system continuously monitors the community and adaptively intervenes. The simulated peer employs two primary mechanisms: (1) it offers heuristic hints to guide learners through unresolved questions without revealing the final answers and (2) generates tailored simulated help-seeking messages to scaffold passive learners into active teaching roles. Furthermore, an empirical study was carried out in a Japanese junior high school to evaluate the system's impact. By embedding these adaptive AI-driven agents within an existing platform, this framework successfully sustains interaction continuity and collaborative knowledge construction as a supplement of human peers.

ABSTRACT. Advances in GenAI-enabled multimodal feedback have shown promising potential for improving learning outcomes. However, empirical research remains limited in examining the factors that influence how students interact with and reflect upon such multimodal feedback, especially in collaborative learning. To address this gap, the present study investigates how groups’ learning characteristics, including language proficiency, learning motivation, and self-efficacy, influence the effectiveness of collective multimodal feedback processing. This exploratory mixed-methods study was based on a self-designed AI-powered vocabulary learning system for Primary 2 Chinese as a second language learners. A total of 83 student participants clustered into 27 groups across 4 classes used the system over a semester. Results suggested that groups’ language proficiency influenced group members’ usage of feedback modalities, and audio-supported content feedback appeared to play a more important role in fostering meaningful interaction and promoting feedback uptake. The findings provide implications for teachers and learning designers regarding the effective integration of multimodal feedback in collaborative language learning.

ABSTRACT. This study examines whether a teachable agent (TA) moderates the influence of mind wandering (MW) tendency on constructive and interactive (C/I) activities in computer-supported collaborative learning (CSCL). A total of 131 university students participated in pairs and were randomly assigned to either a TA or no-TA condition using CoCot, a collaborative concept mapping system. In addition to the TA manipulation, metacognitive suggestions were provided under all conditions as a baseline form of collaborative process support. MW tendencies were assessed using the Daydream Frequency Scale (DDFS), and chat logs from the collaborative phase were coded according to the ICAP framework (Chi & Wylie, 2014). The results indicated that the TA condition produced significantly higher proportions of constructive and interactive utterances and a lower proportion of active utterances than the no-TA condition. Regarding the relationship between MW tendencies and C/I activity, no significant correlations were found in the TA condition, whereas positive correlations between DDFS scores and C/I utterance proportions emerged in the no-TA condition. These findings suggest that a TA exerts an equalization effect on C/I activity—homogenizing engagement, regardless of individual MW tendencies, while also revealing that in the absence of a TA, learners with higher MW tendencies exhibit greater spontaneous generative and interactive engagement. The implications of these findings for CSCL system design and for supporting learners with diverse MW profiles are discussed.

ABSTRACT. Open-ended student comments provide valuable information for quality assurance in higher education. However, transforming fragmented comments into structured reports remains a challenging task. This study investigates the use of large language models to convert student feedback into structured academic quality assurance reports. Specifically, we design and compare five summarization strategies: direct summarization from raw comments, aspect-guided summarization, sentiment-guided summarization, aspect–sentiment-guided summarization, and a hybrid strategy that combines raw comments with structured comment groupings. Experimental results on a Vietnamese student feedback dataset show that direct summarization is a strong baseline and achieves the highest performance on automatic evaluation metrics. Meanwhile, the hybrid strategy obtains comparable results while providing a clearer structure for quality assurance interpretation. These findings suggest that preserving raw comments is important for maintaining the natural context of the input data, while aspect and sentiment information can serve as useful supporting signals for organizing and interpreting generated reports. This study provides an empirical analysis of large language model-based summarization strategies in the context of student feedback analysis and academic quality assurance.

ABSTRACT. Since generative AI is being more widely used to support inquiry, feedback seeking, content generation, and reflection, and since these functions can aid students in handling open-ended tasks in project-based learning (PBL), it is also important to recognize that unguided use of such tools may lead to over reliance. Therefore, this action research study investigates how generative AI can be designed as role-based scaffolding in a high school information technology course. The intervention grouped PBL into four clearly defined phases: project formulation, inquiry integration, creation optimization, and output transfer, and placed generative AI appropriately in each phase as a question assistant, learning partner, design assistant, and reflection partner, respectively. Three action research cycles were carried out with 37 students, and data were collected using a deep learning ability scale. A Depth of Knowledge (DOK)-based cognitive analysis, classroom observations, project artifact evaluations, and interviews were used in this study to investigate students' use of generative AI, revealing clearly that students' use shifted from answer seeking to verification-oriented and reflection-oriented use, and that students made positive gains in critical thinking, creative thinking, self-directed learning, and learning perseverance. More importantly, DOK results showed a definite move from recall and application toward strategic thinking and extended thinking. Thus, the study convincingly argues that generative AI can support deep learning in PBL when its use is mediated by teachers, embedded in task-specific roles, and paired with a critical verification cycle.

ABSTRACT. Artificial intelligence literacy is increasingly essential for primary students, yet age-appropriate pedagogical approaches remain under explored. This mixed-methods study examined whether gamified storybook reading enhances AI literacy compared with storybook reading alone among 87 primary students in Hong Kong. Participants were randomly assigned to experimental (gamified) or control (non-gamified) groups. Quantitative analysis using ANCOVA and Wilcoxon signed-rank tests revealed that storybook reading alone significantly improved knowledge acquisition, while gamification enhanced attitudes, particularly for students with lower initial motivation. However, gamification appeared to interfere with cognitive learning, possibly due to increased extraneous cognitive load from time pressure and leaderboards. Qualitative interviews indicated that students valued both modalities for building interest. These findings suggest that storybooks and gamification serve complementary pedagogical functions, with implications for designing adaptive AI literacy interventions tailored to learner profiles.

ABSTRACT. Computational thinking is an important competence in K-12 information technology education, but its development should not be limited to programming syntax or isolated algorithm instruction. This action research study examined how robotics-supported interdisciplinary collaborative learning could support sixth-grade students’ computational thinking development in a primary school information technology course.

Guided by the 6E learning process, three rounds of classroom activities were designed and implemented: Rock-Paper-Scissors Robot, Fun Helicopter Propeller, and Smart Sunshade. Robot kits, sensors, flowcharts, and block-based programs were used as shared artifacts to support students’ collaborative inquiry, programming, debugging, and reflection. Data were collected through a computational thinking scale, Bebras tasks, classroom observations, student artifact evaluations, and interviews.

The findings showed positive development in five dimensions of computational thinking: decomposition, abstraction, modeling, algorithm design, and evaluation. Classroom observations and artifact evaluations further indicated that students gradually moved from simply completing robot tasks to explaining rules, negotiating program logic, debugging systems, and optimizing solutions around shared artifacts. The study suggests that robotics-supported collaborative learning can make computational thinking more visible, discussable, and revisable in primary information technology classrooms. It also provides design implications for integrating computational thinking development with computer-supported collaborative learning through structured interdisciplinary activities.

ABSTRACT. Generative AI is reshaping teacher professional development. Vibe coding, an approach to developing interactive digital applications through natural language interaction with generative AI, offers new opportunities for educators without programming experience to create educational tools. Nevertheless, teachers still need to learn how to use vibe coding to develop educational tools that effectively meet their instructional needs. This study investigated pre-service teachers’ learning agency, vibe coding behaviors, and learning outcomes in a vibe coding-based educational tool development course, and compared the vibe coding behaviors and learning outcomes between high- and low-agency groups. The results showed that the high-agency group more frequently engaged in behaviors of testing webpages and dissatisfying with the webpage testing results, whereas the low-agency group more frequently checked the textbook. The high-agency group achieved better learning outcomes in terms of the interactive educational webpage quality scores for accuracy of content and support for learning goals. These findings suggest the relationship between pre-service teachers’ learning agency, their vibe coding behaviors, and learning outcomes. The study provides empirical evidence for teacher education and the integration of generative AI into instructional design, helping instructors provide appropriate support based on pre-service teachers’ different levels of learning agency and enhancing their ability to develop educational tools with vibe coding.

ABSTRACT. The proliferation of generative artificial intelligence (AI) in educational settings has intensified concerns around "cognitive delegation"—the habitual outsourcing of higher-order thinking to algorithmic systems. This tendency is particularly consequential during upper elementary school years, when critical and metacognitive capacities are still consolidating, yet existing AI education remains predominantly focused on tool operation rather than critical evaluation of AI-generated outputs. This study proposes and empirically evaluates the SLAP-FACT Matrix Framework, an instructional model integrating two complementary dimensions: SLAP (Source, Logic, Actuality, Perspective) as a multi-layered verification protocol for AI outputs, and FACT (Finding, Attitude, Context, Target) as a structured inquiry scaffold. The framework was implemented through a 10-session program with 23 sixth-grade students in South Korea, assessed via competency checklists and AI interaction log analyses. Finding demonstrated statistically significant improvements across all domains, with students shifting markedly from passive acceptance of AI outputs toward active, multi-dimensional verification and autonomous inquiry. These results offer a replicable, classroom-ready model for cultivating critical AI literacy at the elementary level, with broader implications for embedding intentional inquiry practices across all stages of formal AI education.

ABSTRACT. The rapid adoption of generative artificial intelligence (GEN-AI) in higher education has opened up new opportunities for supporting academic writing. However, many of the existing AI writing tools primarily focus on text generation and surface-level feedback with very little attention to writing development, learner agency and the broader academic literacy needs of international students who struggle in navigating the academic norms in English-dominant higher education environments. Thus, this study presents the design and formative evaluation of a pedagogically grounded AI writing tutor that intends to support international and multilingual learners throughout the academic writing process. The AI writing tutor was developed using a pedagogy first approach informed by the process-genre approach to writing, scaffolding and feedback literacy. Rather than generating complete answers, the tutor guides learners through the stages of planning, drafting, revising, editing and reflection allowing them to retain authorship. To evaluate the tutor’s pedagogical alignment, two Academic advisers with experience in supporting international students participated in an expert review session individually which consisted of tutor interactions, observation and semi-structured interviews. Data was analyzed using thematic analysis, where the findings revealed key themes and provided positive impression of the AI writing tutor’s alignment to pedagogy and framework selected and provided suggestions for improvement that will be implemented in future iterations. This study contributes to the growing field of AI-supported writing by showcasing how educational theory can inform the design of pedagogically grounded AI writing support and provides early formative support for the pedagogical direction of the tutor.

ABSTRACT. Bilingual instruction opens international collaborative learning to more students, but working in two languages is costly: every contribution must be produced twice, may not reach members who do not speak the language used, and is harder to make in a non-native language. This study asks whether avatar-anchored real-time transcription and translation (ARTT) reduces those costs. Eighteen students in a Japanese–English metaverse course compared sessions with and without ARTT, rating it positively for discussion efficiency (M = 4.11, positive response rate 83.3%), understanding others (M = 4.06, 77.8%), and reduced anxiety when speaking (M = 3.89, 77.8%). Gains on the production items were larger for domestic students. Difficulties concerned readability, start-up, and translation accuracy. ARTT may thus preserve what bilingual provision enables while reducing its costs.

ABSTRACT. Recent studies advocate integrating data-driven learning (DDL) and generative AI (GenAI) tools as complementary approaches to addressing EFL learners' challenges in argumentative writing; however, empirical evidence on learner engagement with such integrated platforms remains limited. This study investigated how 31 EFL university students engaged with an AI-powered corpus platform during argumentative writing. Students' behavioural, cognitive, and affective engagement was captured through screen recordings and post-task surveys. Lag Sequential Analysis revealed distinct behavioural sequencing patterns, and cluster analysis identified four engagement profiles: Active Engagers, Efficient Strategists, Self-assured Reviewers, and Reserved Users. The findings offer implications for the pedagogical design of integrated AI–corpus platforms.

ABSTRACT. This study investigates how prospective English-as-a-foreign-language (EFL) teachers developed self-directed learning (SDL) strategies and self-efficacy while using generative artificial intelligence (GenAI) tools to create digital English storybooks. Twelve prospective EFL teachers enrolled in an 18-week graduate course in Taiwan participated in a GenAI-supported materials design project. Data were collected from semi-structured interviews, prompt logs, instructor observation notes, and feedback documents, and were analyzed through thematic analysis. Findings show that participants developed operational, troubleshooting, and learning-related self-efficacy as they learned to use GenAI tools for text generation, illustration design, audio narration, and storybook production. They also enacted SDL through goal setting, resource management, autonomous tool exploration, and emotional regulation. Instructor guidance, peer support, and iterative feedback transformed technical difficulties into confidence-building experiences. The study suggests that GenAI-assisted digital storybook projects can strengthen AI literacy, professional confidence, and lifelong learning competence in teacher education.

ABSTRACT. Classroom videos are valuable for teacher professional learning, but large collections are often difficult to reuse due to challenges in organization and pedagogical retrieval. This paper presents the design of COREference, an AI-enabled classroom video corpus designed to support teacher inquiry through human–AI collaboration. Developed around a corpus of over 3,100 authentic, multi-subject classroom videos, the system transforms raw recordings into searchable, pedagogically meaningful multimodal resources using a two-layer approach. The first layer automates preprocessing via audio filtering, transcript generation, and alignment with expert-defined pedagogical coding schemes. The second layer provides a teacher-facing interface for multimodal querying and AI-supported interpretation of videos, transcripts, and annotations. This paper focuses on the affordances and benefits of the second layer, with initial exploratory feedback indicating that this interaction layer effectively supports teacher inquiries and reflection by streamlining the navigation and retrieval of authentic classroom interactions. Beyond technical processing, COREference demonstrates how an AI-supported corpus can scale professional learning and research reuse without relying entirely on labor-intensive manual review, contributing a sustainable framework for human–AI collaboration in educational infrastructure.

ABSTRACT. Artificial Intelligence (AI) is rapidly becoming infrastructural in education, yet “AI literacy” is often framed as a single competence, masking substantial variation in how AI is positioned in interaction. This scoping review reorganizes the 2020–2026 Large language models (LLM)-in-education evidence base through a role lens. We argue that the same AI can operate as a knowledge utility, dialogic tutor, reviewer, co-creator, teacher assistant, or reflective coach; each role bringing distinct success criteria, risk profiles, and required human competencies. Using targeted database searches, screening, and role-based coding, we synthesize findings across these interaction roles and contribute: (a) an interaction-role taxonomy for educational context, and (b) distinguishing role general co-existence skills from role specific interaction protocols. The paper contributes the notion of role fluency: the capacity to intentionally recognize, select, sustain, switch, and critically govern AI interaction roles in educational activity. The role-fluency lens offers a design and professional development vocabulary for moving beyond generic AI literacy toward accountable human-AI participation in intelligent learning ecologies.

ABSTRACT. For a generative AI tutor to adapt, the learner state must be updated from interpretable response evidence for the preceding question. This study introduces a response-evidence-gated architecture that separates the minimum cognitive demand of an AI question (Q-DOK) from the reasoning demonstrated in a student response (A-DOK), assigning A-DOK only when such evidence is present, and audits the resulting automatic classifications against human judgment. From logs of 37 sixth-grade students in two Korean elementary schools, we derived 999 question–response pairs; in a stratified sample of 200, response evidence was identified in only 37 cases (18.5%). In a conditional comparison sample of 97, exact agreement was 55.7% for Q-DOK (linearly weighted κ = .431) and 75.3% for A-DOK (κ = .560), with over-assignment dominant in both. Among 17 questions rated Q-DOK 3 or higher, 14 responses (82.4%) were A-DOK 1 or 2, supporting the separation of question demand from observed reasoning.

ABSTRACT. In the era of generative artificial intelligence and core competency-driven educational reforms, redesigning assessment poses a critical challenge for teacher professional development. While various technological environments offer distinct affordances for instructional design, how different digital platforms shape teachers’ collaborative reflection and cognitive networks regarding alternative assessments remains underexplored. This study used epistemic network analysis to investigate and compare the cognitive network structures of design feedback from two classes of in-service teachers across two distinct technological environments. The ENA results revealed statistically significant differences between the two groups’ cognitive networks. In summary, while generative AI excels at accelerating individualized technical micro-tuning, an asynchronous knowledge-building environment is essential for driving systemic coherence and community-wide alignment around holistic competencies. This has a crucial implication for future teacher professional development: first, leverage collaborative networks to advance high-level pedagogical concepts, and then harness generative AI tools to execute precise, tactical scaffolding.

ABSTRACT. Improving coding practices such as test-first development is an important goal in programming education. Although reflective learning is known to support improvement, learners often struggle to understand their own coding behaviors and translate reflection into actionable changes. This study proposes a visualization-based reflection and planning approach that supports learners in analyzing their coding activities and formulating plans for improvement. The proposed system visualizes learners’ coding processes using three indicators: test creation timing, test-first compliance, and failed-test persistence. An empirical study was conducted in a third-year undergraduate programming course involving 106 students. Learners engaged in repeated cycles of implementation, testing, reflection, and planning across multiple programming exercises. The results show that learners who engaged in visualization-based reflection and planning demonstrated substantial improvement in coding practices, particularly in test-first compliance, which increased by +18.9%, while learners who did not use visualization showed a decrease of -15.0%. Statistical analysis confirmed that this difference was significant (p = 0.003). However, improvements in other indicators were not statistically significant, suggesting that different aspects of coding practices require different types of support. These findings indicate that visualization alone is insufficient to support behavioral change and that the integration of reflection and planning is essential for improving coding practices.

ABSTRACT. Generative artificial intelligence (AI) is reshaping teaching, learning, assessment, and professional work in higher education. However, institutional approaches to AI integration often remain fragmented. Institutional policies are issued centrally, professional development is delivered as generic tool training, and often teachers are left to decide how AI should be used, disclosed, assessed, and governed in their own courses. The result is a clumsy institutional response to a shared technological transformation, in which educational quality becomes harder to assure, and policy makers, teachers, student support services, and students navigate AI from separate and often uncertain positions. This issue is particularly relevant to tertiary education where programmes are expected to prepare students for organisational roles in which AI increasingly influences analysis, communication, marketing, strategy, and managerial decision-making. Grounded in literature on teacher professional development, AI governance, and assessment design, this paper proposes a Governance-Capability-Curriculum framework for responsible AI integration in tertiary education. The framework links governance, teacher capability, and curriculum and assessment practice to showcase how institutional AI policies can become educationally meaningful. It further contributes a source-critical conceptual model supported by two practical tools: a five-level teacher capability maturity ladder and an assessment assurance cycle for AI-enabled tertiary educational tasks.

ABSTRACT. This study examines how middle-school science teachers collaboratively address linguistic and cultural diversity through practice-based micro-Communities of Practice in Indian science classrooms. Drawing on Communities of Practice and micro CoP theory, we analyzed semi-structured interviews with 12 teachers from two urban schools. The findings identified three practice-based micro-CoPs: multilingual science language scaffolding, culturally contextualized science representation, and diagnostic participation-support planning. These micro-CoPs were organized around recurring pedagogical concerns rather than fixed stakeholder groups. Across the three micro CoPs, teachers worked with colleagues, parents, counsellors, school leaders, and external experts to make science concepts more accessible, culturally meaningful, and responsive to students’ participation barriers. The collaborations generated shared practices such as bilingual cue cards, visual word banks, localized examples, culturally sensitive discussion routines, barrier checklists, modified worksheets, home-support routines, and follow-up communication practices. Digital and material artifacts, including WhatsApp groups, home-language videos, online records, worksheets, models, and checklists, supported the movement of these practices across classroom, home, counselling, leadership, and community contexts. The study contributes a practice-centered account of how localized teacher collaborations generate, adapt, and circulate reusable repertoires for language- and culture-responsive science teaching in diverse classrooms.

ABSTRACT. Large-scale asynchronous online ICT literacy courses require learners to acquire diverse skills in a stepwise manner. We have been improving our courses by incorporating Keller’s Personalized System of Instruction (PSI) model. In previous implementations, PSI-based mechanisms, including block assignments, mastery requirements, and opportunities for resubmission, were introduced. However, despite these improvements, the course completion rate declined from 95.9% to 93.6%.

Within the PSI framework, proctors play a critical role by monitoring learners’ progress and providing appropriate guidance and feedback. In large-scale asynchronous online courses, however, it is difficult for instructors to provide individualized feedback to every learner. To address this issue, we introduced a peer-review mechanism as a means of supplementing the proctor function.

The purpose of this study is to investigate the effects of peer review on learners’ learning behaviors and course completion rates. Through peer review, learners are expected to gain greater awareness of their level of achievement. It is hypothesized that learners who receive low peer-evaluation scores will increase their engagement with learning materials and revise and resubmit their work accordingly. As a result, the introduction of peer review is expected to contribute to an improvement in the overall course completion rate.

ABSTRACT. In engineering education, learning activities often involve students working in the physical laboratory space and in groups. While recent advances in multimodal artificial intelligence and social robotics create new opportunities to support hands-on learning in science and engineering laboratories, many learning analytics platforms still rely primarily on digital trace data, limiting their ability to capture embodied, physical, and collaborative learning processes. This paper presents the design and pilot implementation of authoring multimodal robot-mediated interactions within the LA-ReflecT learning platform. The proposed approach integrates learning and robot management functions, learner interaction logging, and sensor data (like camera) to analyze robot-supported feedback possibilities and create a human-centered learning environment that connects physical task performance with reflective learning support. The pilot study, focusing on a circuit-building laboratory activity, involved 20 learners working in groups of 4 for a 3-hour lab session. The platform was used simultaneously by 2 groups, with the robot serving as an observing peer and collecting snapshots of observed task progress. A machine learning-based classifier was then used to tag each observed frame to understand human-object interactions. The pilot demonstrates the feasibility of embedding multimodal sensing and robot interaction into LA-ReflecT to support learning activities distributed in physical space. With the multimodal learning analytics, a design framework and initial implementation approach provided insights for extending the platforms toward physical, collaborative, and AI-supported laboratory learning contexts. Future work will refine the interaction model, improve automated interpretation of circuit-building actions, and evaluate the impact of robot-supported reflection on learners’ conceptual understanding and problem-solving behavior.

ABSTRACT. This study explored whether realistic drawing performance and spatial ability could predict 3D design performance among design students. A total of 79 undergraduate students from a design-related department participated in the study. Participants completed a spatial ability test, a realistic drawing task, and a 3D design project. Descriptive statistics, Pearson correlation analysis, and multiple linear regression were used to analyze the data. The results indicated that both realistic drawing performance and spatial ability were significantly and positively associated with 3D design performance. Regression analysis further showed that both variables independently predicted students’ 3D design outcomes. Interestingly, no significant correlation was found between realistic drawing performance and spatial ability, suggesting that they represent different but complementary abilities in the design learning process. Descriptive findings revealed that most students demonstrated moderate levels of performance across the four dimensions of 3D design assessment, whereas only a limited number achieved advanced performance levels. The findings imply that observational drawing skills and spatial cognitive abilities continue to play meaningful roles in students’ 3D design learning and should be considered when developing instructional strategies for design education. The study provides practical implications for educators by highlighting the value of incorporating both drawing-based observation training and spatial ability development into 3D design courses.

ABSTRACT. This study applies conjecture mapping to pedagogist competency development in early childhood education (ECE). Through qualitative interviews with three educational consultants at a multi-site ECE organization in Ontario, Canada, and systematic coding of 160 segments across 20 codes, three iterative conjecture maps are developed. Findings identify six design principles, three theoretical conjectures, and four design constraints governing pedagogist competency. The most consequential unanticipated mediating variable, the artist-inquiry disposition, is identified as a high-leverage competency not captured in any existing ECE framework. Implications for pedagogist professional development and self-assessment tool design are discussed.

ABSTRACT. This work-in-progress poster presents our preliminary work on a pedagogical framework, Intelligent-supported Authentic Inquiry-driven Design Learning (IAIDL), that explores how to leverage AI and GPS technologies to support outdoor inquiry learning in design education, with the aim of enhancing their perceptual abilities, authentic inquiry competence, and iterative design thinking. The paper discusses the rationale and theoretical grounds (a crossover between inquiry-based learning and design-based learning) of IAIDL, as well as how it can be operationalized in a market research course in design education.

ABSTRACT. While peer assessment has been widely adopted in technology-enhanced learning (TEL), limited research has examined how learners' psychological dispositions shape peer-feedback behaviors in multimodal contexts. This study investigates the relationship between academic hardiness and peer-feedback engagement in a video-based TEL environment. Fifty-three undergraduate students in an Environmental Communication course participated in three rounds of anonymous peer assessment supported by a digital platform with timestamped annotations and rubric-guided feedback. Academic hardiness was measured across four dimensions (commitment, control-effort, control-affect, and challenge), and peer-feedback comments were coded into affective, cognitive, and metacognitive categories. Spearman's rank-order correlations and Mann-Whitney U tests were conducted. Results revealed selective relationships between academic hardiness and feedback types. Commitment was positively associated with affective feedback, whereas challenge was positively associated with cognitive feedback. No significant associations were found for metacognitive feedback. These findings suggest that academic hardiness supports emotional and cognitive engagement but does not automatically lead to higher-order reflective feedback. The study contributes to TEL research by demonstrating that learner dispositions may shape peer-feedback engagement in multimodal learning environments, highlighting the potential role of instructional scaffolding in supporting metacognitive feedback.

ABSTRACT. While large language models (LLMs) are increasingly utilized for automated grading, their performance remains highly sensitive to task type, metric selection, and prompt architecture. We address the technical bottleneck of inconsistent STEM assessment by evaluating numerical scoring across three disciplines (physics, chemistry, and mathematics) using five widely-used LLM models (GPT-4o, Gemini 2.0 Flash, Claude 4.0, DeepSeek V2.5, and Grok 4) under two distinct prompting regimes: rubric-plus-exemplar and exemplar-only. Our analysis reveals that Claude 4.0 (without a rubric) excels in physics and chemistry, whereas Grok 4 (with a rubric) leads in mathematics. Crucially, exemplar-only prompting consistently yields higher exact-match rates and greater distributional overlap than rubric-inclusive prompts, suggesting a "context saturation" threshold for complex rubrics. Although subject-specific ensembles provide marginal (+1.5%) gains, Grok 4 is the sole model to significantly reduce AI–human bias below the human–human baseline. Ultimately, these findings establish an empirical foundation for deploying LLMs as decision-support tools in STEM education and offer actionable guidelines for prompt optimization.

ABSTRACT. The current study investigated the impact of incorporating reflective learning and generative AI tools in Taiwanese high school writing classes on students' self-efficacy, engagement, and writing performance. The study involved 57 twelfth-grade students from an urban high school, divided into three instructional groups: AI-integrated learning(Control), reflective journaling (Experimental Group A), and a combination of AI-integrated and reflective journaling (Experimental Group B). Pre-course and post-course assessments measured outcomes using IELTS Task 1 writing tests, with data analyzed via one-way ANCOVA and thematic analysis. Results revealed that the AI-only control group maintained high self-efficacy and showed the highest behavioral engagement. Meanwhile, the combined group experienced a statistically significant decline in writing self-efficacy. There were no significant differences in writing performance across groups due to the short intervention. However, qualitative logs indicated the combined group successfully developed metacognitive awareness of their linguistic limits. In conclusion, the self-efficacy actually helps students to be more conscious of their learning limitations. The study highlights the importance of a phased pedagogical approach to AI integration, demonstrating that AI tools should be carefully balanced with metacognitive reflection to prevent an illusion of competence and foster genuine self-regulation.

ABSTRACT. In view of the positive learning effects of learner-centered error-correction activities and the prevalence of printed error-correction books, this work targets an online error-correction notebook automatically generated by the system, and examines students’ perceived benefits of this newly developed component for the support of Chinese language learning. A class of fifth-grade students (n=24) participated for six consecutive weeks. Results indicated that a predominant percentage of the students (n=22) affirmed the augmented feature as better supportive of their learning than the original (i.e., no system auto-compilation). Additionally, a chi-square test indicated that students’ selections between the two versions differed significantly, χ²(1) = 16.67, p < .001. Furthermore, thematic analysis highlighted the enhanced function’s usefulness in transforming the error-correction task into a more organized, efficient, and targeted review process while heightening error awareness.

ABSTRACT. This study integrated an in-house AI writing system with the 8P theoretical framework to develop an effective instructional approach aimed at enhancing college students’ creativity and English story writing proficiency. 32 college students participated in this study. Data were collected using pre- and post-creativity scales and English story-writing tests. The results demonstrated significant improvements in students’ creativity and substantial improvement in English story-writing proficiency. The 8P instructional approach in AI-supported learning provided clear learning goals and structured environmental demands that encouraged the students to apply their creativity in completing original digital picture books. During mental operations for producing stories in English, students received AI support that helped them brainstorm ideas, select appropriate words, check for grammatical errors, and refine story structure. These findings suggest that integrating the 8P instructional approach with AI-supported features can effectively foster college students’ creativity and enhance their writing proficiency.

ABSTRACT. In the smart era, teachers' online teaching ability has attracted much attention, and how to evaluate it objectively and effectively has become a key issue to promote development. Existing studies on teachers' online teaching ability evaluation rely on video analysis and peer observation, lacking the student perspective. This study proposes an intelligent evaluation method based on students' learning experience using pre-trained models. We construct the SOLE dataset (12,708 manually coded student comments from Chinese university MOOCs) and build a text classification system. Experiments show that the BERT-CNN model achieves optimal performance with 90.2% accuracy and F1 score of 0.8918. Based on the classification results, we provide personalized teaching ability portraits and comparative analysis across different MOOC courses, along with directions for future research.

ABSTRACT. Recent studies in Computing Education Research have highlighted the importance of cognitive processes such as code comprehension, tracing, debugging, and problem decomposition in introductory programming education. Meanwhile, English as a Foreign Language (EFL) instruction in higher education frequently engages students in activities that require structural understanding and analytical processing, including academic writing, reading comprehension, syntactic parsing, and revision. However, little attention has been paid to the possible relationship between skills fostered through foreign language learning and programming readiness among novice programmers. Drawing on findings from research on novice programmers and code comprehension, this paper explores potential connections between learning activities commonly found in EFL instruction and competencies associated with programming readiness. A conceptual framework is proposed to identify areas of overlap between academic writing, reading comprehension, syntactic parsing, rereading and revision, task interpretation, and cognitive processes emphasized in introductory programming education. This paper does not claim a direct transfer from foreign language learning to programming ability. Rather, it proposes a conceptual framework for examining how structural and analytical literacy developed through EFL instruction may relate to programming readiness and suggests directions for future empirical research.

ABSTRACT. This study investigates the spatial and temporal characteristics of facial expressions to classify academic emotions—boredom, confusion, engagement, and frustration—using the DAiSEE dataset. Two frame selection methods, Targeted and Changepoint, extracted three frames from each 10-second video, and deltas (Euclidean distance, cosine similarity) captured temporal changes in facial landmarks. Six feature selection techniques, including insights from teacher interviews, were tested with five machine learning algorithms: K-Nearest Neighbor (KNN), Decision Tree (DT), Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Support Vector Machine (SVM). The findings suggest that Changepoint frame selection has a slight edge in classifying emotions involving dynamic expressions, such as engagement and frustration, whereas the Targeted method performs well across all emotions, especially more static ones like boredom and confusion. Interestingly, models trained on features from only the left side of the face performed comparably to those using features from the entire face, underscoring the relevance of specific facial regions in emotion classification. Additionally, these experiments validated teachers’ insights regarding which parts of the face typically reflect academic emotions. Whole feature selection generally excelled in boredom and engagement classification, while Teacher and Left feature selections were most effective for confusion and frustration, respectively. KNN and CNN consistently outperformed other algorithms, with KNN being most effective for boredom and CNN for confusion, engagement, and frustration. These insights underline the potential of integrating temporal dynamics and feature selection to improve academic emotion classification accuracy.

ABSTRACT. This study examines how emotional granularity and task-fit of large language model-based chatbot responses affect user trust. We propose a Conditional RAG framework that controls emotional expression by combining prompt-based emotion type specification with VAD-based retrieval of emotionally similar examples. Using this framework, chatbot responses were generated under seven conditions: joy (high/low), sadness (high/low), anger (high/low), and neutral. An online mixed-factor experiment with three scenarios and 175 participants was conducted to evaluate the trust. Significant condition effects were observed in all the scenarios. Task-congruent emotional responses generally increased trust compared to incongruent and neutral responses. However, a higher granularity is not always beneficial. In negative contexts, low-granularity responses sometimes produced higher trust than high-granularity responses, suggesting that restrained empathy may be more effective than intense emotional expression. These findings indicate that the emotional design of chatbots should prioritize contextual appropriateness rather than simply increasing expressive richness.

ABSTRACT. This study examines media balance in inquiry learning using log data collected from RefNavi, a web-based platform for information collection and management. The dataset consisted of 15,378 information resources registered by 2,189 student users. Users were categorized into four groups according to the number of registered resources, and differences in media-use patterns were analyzed using chi-square tests and residual analyses. The results revealed significant differences among activity groups. Highly active learners used a wider range of media, including web resources, statistical data, reports, videos, and academic papers, whereas low-activity learners relied more heavily on academic papers and interviews. Differences were also observed in monthly registration patterns. These findings suggest that diversity of information sources, rather than reliance on a particular media type, may be associated with extensive information-gathering activity in inquiry learning.

ABSTRACT. While Large Language Models (LLMs) have vast knowledge, they often struggle to adapt their teaching style to individual students, treating a struggling learner the same as an advanced one. Motivated by the need for more adaptive and resource-efficient AI tutors, this study compares a standard monolithic Single-Agent workflow against a specialized Multi-Agent workflow. Using the Cambridge IGCSE Economics curriculum, we generated responses for 60 questions across four simulated student personalities. We compared a single 70B-active-parameter model (Llama 3.3 70B) against a chain of three smaller 17B-active-parameter models (Llama 4 Scout) to test if architectural specialization yields better results than raw size. Blind evaluations by both AI systems and human educators indicate that the 70B Single-Agent model consistently outperformed the Multi-Agent configuration in terms of Accuracy [+1.05], Personalization [+2.04], and Pedagogical Quality [+2.02], despite similar computational budgets. However, the Multi-Agent workflow proved to be more computationally efficient, using just 39% of the Single-Agent’s computational cost. These findings suggest that, for K–12 tutoring tasks of this scope, overall model capacity plays a more decisive role than agent specialization, and that multi-agent architectures may not inherently yield pedagogical benefits without sufficiently strong underlying models.

ABSTRACT. The generative artificial intelligence (GenAI) tools offer clear benefits for learning and administrative efficiency for both students and educators in higher education. However, GenAI tools can also pose serious risks to patient confidentiality, professional accountability, and data governance when used inappropriately. Hence, the study examines the ethical boundaries surrounding the use of GenAI in clinical and practice placement settings within health and social services education in New Zealand Polytechnics. As such, despite the importance of having clear guidelines on the ethical use of GenAI, there remains a lack of specific policies governing its use in clinical and practice placement settings. While professional standards policies exist, they do not explicitly address GenAI use in detail. Health New Zealand has also issued precautionary guidance restricting the use of GenAI in a clinical context, but these policies primarily focus on prohibitions and provide limited practical direction on the ethical use of GenAI in practice placement settings. Therefore, the study adopts a conceptual policy analysis approach to examine existing guidance on GenAI use and identify gaps in the need for student-focused, placement-specific guidance. In addition, the study introduces a conceptual framework called the clinical GenAI boundary model, which provides explicit guidelines with real-life examples to support students in navigating the ethical use of GenAI within health and social service contexts.

ABSTRACT. This study addresses the lack of engaging and age-appropriate instructional materials in marine sustainability education for elementary learners. Although environmental education often emphasizes topics such as plastic pollution and climate change, sustainable seafood concepts remain underrepresented and difficult for younger students to understand because they involve abstract ideas such as overfishing, resource depletion, seasonal consumption, and responsible decision-making. To address this issue, this study develops a game-based learning (GBL) digital system that integrates interactive storytelling, animation, feedback, and scenario-based tasks to support students’ understanding of sustainable seafood and marine environmental issues. The system is structured into four modules covering fishing methods, seasonal seafood knowledge, and sustainable consumption principles, guiding learners from conceptual understanding to real-world application. A mixed-methods approach is adopted, involving 30 elementary school students, pre- and post-tests, an ARCS-based learning motivation questionnaire, and a Technology Acceptance Model questionnaire. Descriptive statistics are used for data analysis, and inferential statistical methods are not applied, as this study is positioned as a pilot study aimed at obtaining preliminary insights for system refinement rather than statistical generalization. The findings suggest that the GBL system improves students’ conceptual understanding, learning motivation, and system acceptance. These findings will be used to further improve the game design and to support future research with larger samples and inferential statistical analysis. This study contributes to the design of game-based digital learning environments for sustainability education and provides practical implications for integrating GBL into environmental education for young learners.

ABSTRACT. Game-based assessment (GBA) has demonstrated considerable potential for supporting formative assessment; however, most existing scaffolding designs remain static and predefined, limiting their ability to address individual learners' diverse needs. This pilot study integrated a generative AI-based dynamic scaffolding feature as an AI mentor into an existing massive online card-based educational game GBA platform to examine its preliminary effectiveness. Twenty-one fifth-grade students engaged in a 40-minute session with the AI mentor activated. Quantitative results indicated that students demonstrated high technology acceptance and low activity anxiety and cognitive load toward the AI mentor. Qualitative coding revealed that most students perceived the AI mentor as helpful in providing hints, expressed willingness to interact with it, and demonstrated general trust in its responses. These findings provide preliminary support for the feasibility of integrating generative AI-based dynamic scaffolding within GBA contexts and inform the design of future experimental studies.

ABSTRACT. The role of smartphones in schools has been widely debated in recent years, and this study foregrounds students’ voices on how smartphones shape learning ecologies in both classroom and self-directed study contexts. Data were collected from 57 students in two Finnish lower secondary schools through open-ended questionnaires. Drawing on Laclau and Mouffe’s discourse theory, we examined how meanings around smartphone use are constructed through articulation. The findings showed that smartphones were used both as teacher directed tools and as flexible, student driven resources. According to students, phone use was mainly guided by teachers for information retrieval and quiz based activities, while students also independently used their phones for a wider range of purposes, including watching instructional videos, learning via social media, using AI for explanations, and handling practical tasks such as note taking. At the same time, students described several tensions related to smartphone use, including the dual role of phones as both learning resources and sources of distraction, the trade off between fast information access and its perceived unreliability, and mismatches between students’ practices and teachers’ rules. This suggests a need for clearer classroom practices that encourage more purposeful smartphone use while ensuring it supports, rather than disrupts, learning.

ABSTRACT. In self-regulated learning (SRL), planning is crucial for achieving goals, yet many learners struggle to create effective plans. Deciding “what to learn” is particularly challenging because it requires an objective assessment of knowledge states and decision-making among numerous options. This study developed and evaluated a learning-planning support dashboard that combines knowledge-state visualization using the Open Knowledge and Learner Model (OKLM), which estimates learner knowledge states by linking learning-activity logs to knowledge models, with recommendation features. We conducted an empirical study during a two-week winter break. The proposed dashboard provided visualizations of unit-level understanding, detailed planning, recommendations for learning units and resource types, and monitoring and reflection functions. Results showed that learners who engaged in monitoring and reflection had significantly longer study times than those who did not, suggesting that the dashboard may have helped sustain engagement in learning. However, changes in SRL skills were not significantly associated with short-term gains in proficiency. The recommendation feature had low usage rates and showed no significant effects on learning outcomes, partly because the knowledge-state visualization itself may have provided sufficient decision-making support. Qualitative analysis of free-response comments supported this interpretation, with learners reporting that the knowledge map helped them identify what to study at a glance. These results suggest that OKLM-based knowledge-state visualization supported learners' “what to learn” decision-making, which was associated with greater engagement during the self-study period, while further design improvements are needed to promote effective use of recommendation features.

ABSTRACT. Generative Artificial Intelligence (GenAI) can be misused to disguise plagiarised programming submissions on assignments, allowing GenAI use. The disguises tend to be pervasive, while the solutions might share GenAI-inspired program flow. Hence, we present a plagiarism detector that can handle pervasive changes using cosine similarity and focuses on token occurrences to ignore shared program flow. The detector is dedicated to data science assignments, where only identifier names unrelated to data science library names are generalised to handle identifier renaming. According to our evaluation of 350 GenAI-disguised copied submissions and 1,417 non-copied submissions, the plagiarism detector is somewhat effective. It can identify 78% similarity in copied submissions despite the complexities of some changes (e.g., statement reordering and changes in the control structure). Further, it achieves 80% top-5 precision. Compared to a baseline approach, the detector is slightly more effective. Pervasive changes by GenAI tend to substantially reduce the resulting degree of similarity.

ABSTRACT. Scratch lowered syntactic barriers and invited young people into creativeprogramming; generative AI now lowers code-production effort further. This conceptualpaper asks how programming education should respond when articulating goals,interpreting information, and evaluating generated behavior may become moreconsequential than producing code. We position systems thinking (ST) as anorganizing lens—not a substitute for programming—for making intended wholes,boundaries, and relationships explicit. Goal-to-Milestone implements a system-first,evidence-returning relational refinement cycle: learners maintain a project-systemmodel, narrow it to one relationship and observable completion signs, then build,preview, judge, and revise. A structured conceptual synthesis and prototype coherenceaudit informed the architecture. Relation-level scoping may shorten the path to aninterpretable preview while preserving learner authority; the framework remainsunevaluated.

ABSTRACT. Generative AI music tools (e.g., Suno, Udio) have significantly lowered technical barriers to music creation, stimulating broad societal discourse on “technological democratization.” This study analyzes 15,406 valid danmaku (bullet-screen comments) collected from Bilibili to investigate public cognitive trajectories and their implications for music aesthetic education. Employing NLP, K-Means clustering, and temporal sentiment evolution modeling, we identify four key findings: (1) discourse exhibits a markedly uneven four-dimensional structure — technology spectacle dominates (54.4%), followed by evaluation of specific musical elements (33.3%), with ethical controversy (6.3%) and traditional aesthetic resistance (6.0%) at the margins; (2) public sentiment follows a three-stage logic spanning a latent period (2017–2023), a Suno-driven explosive growth period (2024), and a differentiation period marked by aesthetic reflection and ethical anxiety (2025–2026); (3) a fundamental paradox exists within democratization discourse — tool accessibility has not eliminated the knowledge gap in aesthetic evaluation, revealing a decoupling of technical access from aesthetic democratization; (4) platform interactions have spontaneously formed a “danmaku shadow classroom” facilitating informal music knowledge transmission. Based on these findings, we argue that contemporary music aesthetic education urgently requires repositioning: its core should shift from traditional “creative practice” toward cultivating “critical listening” and “aesthetic metacognition,” while formally integrating cross-media aesthetics and AI ethics into educational frameworks.

ABSTRACT. As generative AI (GenAI) increasingly automates code production, programming education faces a critical question: how is computational thinking (CT) enacted when learners no longer write every line of code themselves? This study introduces AI orchestration—learners’ active process of articulating computational intent, directing GenAI outputs, evaluating generated artifacts, and refining prototypes—as a lens for reconceptualizing CT in GenAI-integrated programming education. A four-week course grounded in Design Thinking was implemented with 18 gifted middle school students in Computer Science, combining Gemini for Education with a Python-based web application framework (Streamlit). CT was assessed with the Computational Thinking Scales (CTS) before and after the course, and Wilcoxon signed-rank tests were conducted on CT and its five sub-components. CT increased significantly from pretest to posttest (p = .007, r = .69), with large effects observed for Algorithmic Thinking (p = .004, r = .76), Creativity (p = .011, r = .68), and Critical Thinking (p = .045, r = .64). No significant changes were detected for Problem Solving (p = .639) or Cooperativity (p = .128). These patterns suggest that CT in GenAI-integrated environments is enacted less through manual code production and more through the orchestration of GenAI particularly through articulating computational intent, evaluating outputs, and iteratively refining artifacts. The findings provide preliminary evidence that CT might be reconfigured around AI orchestration, and they point to the need for assessment tools that capture learners’ AI orchestration in GenAI-integrated programming education.

ABSTRACT. This exploratory study examined whether technology-assisted music training can facilitate L2 speech perception and production, with particular attention to pitch-related prosodic processing. Building on previous findings that Japanese learners of English (JLEs) experience perceptual difficulty when pitch shifts from a lower to a higher register (male→female speaker order), the study investigated whether music training involving alternating male and female vocal parts could improve F0-based speech processing. Four Japanese university students with different levels of musical aptitude participated in a pre-test/training/post-test design. Training consisted of singing practice using a karaoke application with real-time acoustic feedback. Participants completed an AX discrimination task and a speech production task before and after training. Production samples were analyzed acoustically for pitch range and speech duration and were evaluated by native English listeners for comprehensibility and accentedness. Results showed limited but consistent improvements in comprehensibility and perceptual accuracy for upward pitch shifts. Item-level analyses further suggested that discourse-level prosodic encoding remained difficult even for participants with high musical aptitude. These findings tentatively suggest that while short-term music training may facilitate acoustic and perceptual aspects of L2 prosody, discourse-level prosodic control may require additional abilities beyond accurate pitch processing alone.

ABSTRACT. This design-oriented paper presents TextQuest AI, a teacher-oriented authoring environment for creating short AI-NPC storyworld activities for text exploration. The system addresses a practical design problem: teachers may wish to transform readings, local-cultural materials, historical cases, or inquiry topics into interactive learning experiences, but they often lack time, technical skills, and narrative-design expertise. TextQuest AI supports teachers in converting source texts into locations, AI-powered non-player characters (NPCs), clues, tasks, and reflection prompts. In the student interface, learners enter a web-based game-like environment, interview AI-NPCs, collect evidence, compare perspectives, and produce a final interpretation within an approximately 30-minute activity. In the teacher interface, an AI design assistant helps analyze texts, suggest NPC roles, generate dialogue rules, propose tasks, and check alignment with learning goals. The initial evaluation will involve 26 teachers. After a 20-minute explanation and guided trial, they will use the system for 40 minutes to design an AI-NPC storyworld activity and then evaluate its usability, usefulness, pedagogical value, teacher control, and classroom feasibility. The paper describes the design rationale, system workflow, example scenario, and evaluation plan.

ABSTRACT. With the implementation and advancement of China's digitalization strategy, the digital transformation of higher education is becoming an important driving force for urban-rural integration and development. Based on the panel data of 31 provinces in China from 2011 to 2023, this study constructs an econometric model to empirically examine the impact of the digital transformation of higher education on urban-rural integration and development, and depicts its spatiotemporal evolution characteristics. The research findings are as follows: (1) The digital transformation of higher education and the level of urban-rural integration and development in China are on the rise, but the issue of spatial injustice and imbalance has become increasingly prominent; (2) The digital transformation of higher education significantly promotes urban-rural integration and development, with the shift in teaching and research, as well as human modernization, being of primary importance. Digital support and scale demonstrate weak effects due to not touching the core; (3) The promoting effect of the digital transformation of higher education exhibits heterogeneity in time and space. On the one hand, its benefits rely on the improvement of infrastructure and macro policy guidance, and are influenced by the interplay of major public events and economic and social fluctuations. On the other hand, the connotative drive of high-end talent and industrial governance in the eastern region has become an ecological reconstruction force. The digital infrastructure and talent release in the central region yield higher marginal benefits but are limited by the gap in local innovation capability and achievement transformation efficiency. The western region, relying on external resources and policy tilting, has achieved significant results but faces the challenge of fragile endogenous systems. Therefore, suggestions are proposed from the dimensions of strengthening overall ecological investment in the digital development of higher education, promoting different digital strategic actions according to local conditions, establishing relevant policy guarantees and monitoring and evaluation.

ABSTRACT. The increasing integration of generative artificial intelligence tools such as ChatGPT in higher education has reshaped how learners approach academic tasks and learning processes. This study explores how ChatGPT influences trainee teachers’ learning experiences through the lens of Self-Determination Theory (SDT) and basic psychological needs. Specifically, it examines how the use of ChatGPT shapes learners’ sense of autonomy, competence, and relatedness in their academic activities. This study adopted a qualitative phenomenological approach, involving semi-structured interviews with seven trainee teachers who actively use ChatGPT for learning purposes. The data were analyzed using thematic analysis to identify patterns in participants’ experiences. The findings indicate that ChatGPT functions as an academic support tool that enhances learners’ autonomy, competence, and relatedness in learning processes. Participants reported greater autonomy over their learning, a stronger sense of competence in completing academic tasks, and new forms of supportive interaction that shaped their sense of relatedness in learning. However, the findings also suggest that excessive reliance on ChatGPT may influence the depth of learning and reshape interpersonal interactions in learning contexts. Overall, the study highlights that ChatGPT plays a significant role in supporting learners’ sense of autonomy, competence, and relatedness while also presenting challenges that require balanced and critical use. The findings contribute to ongoing discussions on artificial intelligence in education and provide insights for the effective integration of AI tools in higher learning environments.

ABSTRACT. With the proliferation of online learning, shallow reading which characterized by skimming and brief glances has become a common daily learning strategy. However, accurately identi-fying shallow reading poses a significant challenge. Concur-rently, eye-tracking technology, as the optimal feedback of psychological activity, emerges as one solution. Yet in studies using eye movements to identify student reading states, the selection of static features has limited the accuracy of shallow reading recognition. To enable page-level identification of shallow reading, this study proposes a shallow reading recog-nition model based on a DistilBERT-based encoding frame-work for multimodal data. This system builds upon multi-modal data for learning behavior recognition. Specifically, for identifying shallow reading, it not only relies on the I-VT al-gorithm to extract gaze, saccade, and fixation events from raw eye-tracking data but also leverages timestamps from ebook data to inject spatial information about these events. The trained EyeFormer model achieved an AUC of 0.963 ± 0.015 and an accuracy rate of 90.1% in a page-level task during a controlled experiment involving 63 non-native English-speaking university students. These results demonstrate the model's high precision, providing a reliable behavioral foun-dation for subsequent personalized reading interventions based on generative AI.

ABSTRACT. Today’s generative AI systems increasingly produce hallucinated or flawed representations in learning contexts, yet such outputs are typically treated as risks to be minimized. Grounded in AIED and learning sciences, this study examined whether hallucinated AI artifacts can instead function as productive resources for collaborative sensemaking. We report a quasi‑experimental classroom study in higher education in which student teams engaged with either hallucinated or canonical versions of a domain model prior to instruction. Using mixed‑methods analyses, we traced how early sensemaking, model revision, and collaborative understanding unfolded over time. Results indicate that exposure to hallucinated artifacts initially afforded confidence and surface-level coherence, but also introduced epistemic uncertainty that prompted questioning, critique, and revision during collaboration. Although these artifacts did not support early correctness, they resulted in deeper model‑based reasoning, stronger evaluative transfer, and more flexible epistemic stances. In contrast, canonical artifacts supported early convergence but encouraged more verification‑oriented reasoning. Interpreted via learning theories of productive failure, cognitive conflict, and epistemic vigilance, these findings suggest that AI hallucinations, when intentionally designed and pedagogically scaffolded, can function as failure‑driven epistemic perturbations. We briefly discuss implications for AIED system designs that prioritize sensemaking, epistemic agency, and responsible human-AI collaboration over short‑term accuracy.

ABSTRACT. Drawing on basic psychological needs theory (BPNT: competence, autonomy, relatedness), we explored 20 EFL undergraduates’ need satisfaction when using an AI‑powered corpus‑based writing platform (AI‑Corpus Writing). Using Q methodology, analysis of 20 Q‑sorts revealed four distinct basic psychological needs (BPNs) satisfaction profiles: (1) cautious users preserve autonomy by directing AI suggestions rather than being directed; (2) relational resistors satisfy competence and relatedness almost exclusively through human feedback, rejecting AI’s social utility; (3) pessimistic dependents exhibit a triple BPN deficit (low autonomy, low competence, absent relatedness), passively copying AI outputs; (4) skill‑empowered optimists reframe AI proficiency as a higher‑order competence, maintaining autonomy by filtering suggestions. Across all profiles, learners agreed that AI‑corpus suggestions are reliable and expand expression, but excessive suggestions overload decision‑making, directly threatening autonomy. We conclude that psychologically calibrated AI writing tools must limit suggestion overload, preserve learner agency, and leave room for human feedback, or risk widening BPN‑based learning disparities. The findings are expected to inform pedagogical practices and designs for building psychologically supportive AI-supported writing environments.

ABSTRACT. As the necessity of text-based programming education for elementary students grows, learners often experience severe cognitive overload and demotivation during the transition from visual block-based environments to text-based languages like Python. Furthermore, traditional programming education frequently relies on abstract, contrived examples, making it difficult for students to perceive the practical value of coding and often neglecting the cultivation of Social-Emotional Learning (SEL) competencies essential for contemporary education. To address these challenges, this study proposes a 12-session instructional design for a "Smart Class Agent" based on a "Data-Logic Fusion" approach, targeting fifth-grade students. This pedagogical model meaningfully integrates authentic, real-world classroom data—such as school event schedules and peers' facial expressions—with core logical control structures. To mitigate the cognitive burden of unfamiliar syntax, the program systematically applies cognitive scaffolding strategies, including 1:1 structural contrast and template-based tinkering. Through this contextualized learning experience, students collect and manipulate data relevant to their daily lives to build a modular agent that resolves actual classroom inconveniences and provides emotional support to peers. The anticipated outcomes suggest that the Data-Logic Fusion model will not only facilitate a seamless, low-anxiety programming language transition but also simultaneously enhance learners' computing thinking (CT) by grounding algorithms in reality, and foster their SEL competencies by empowering them to contribute positively to their classroom community

ABSTRACT. Extensive research has examined e-learning adoption using the Technology Acceptance Model (TAM). The most relevant TAM elements for e-learning are computer self-efficacy, subjective/social norms, perceived enjoyment, and system quality, according to the literature. In this study, we developed a comprehensive TAM model that considers student privacy. We aimed to understand students’ perspectives on data privacy in higher education through a survey at a Japanese university (N=286). A theoretical TAM model was used and expanded to survey the students. The hypothesis tested through structural equation modeling linked students’ behavioral intention to use EdTech with their privacy perceptions. Based on these findings, we proposed research, design, and policy recommendations to protect student privacy in educational settings.

ABSTRACT. This study explores whether gamified chatbot instruction improves primary students’ engagement in different modes. A randomized controlled trial compared individual, collaborative, and traditional groups, using school engagement scales and interviews. Results showed a significant time × group interaction for overall engagement. Emotional and cognitive engagement improved in both gamified groups, while the control group declined. In conclusion, gamified chatbot can effectively enhance engagement, offering insights for designing engaging language instruction

ABSTRACT. Collaborative learning is a complex socio-emotional process. A lack of effective regulation mechanisms often leads to collaboration failure; therefore, Socially Shared Regulation of Learning has become a frontier of research in the learning sciences. This study adopted a design-based research methodology, aiming to explore how to support shared emotion regulation in collaborative learning through the design of specific emotion scaffolds. The results indicate that emotion regulation scaffolds can effectively promote the teams’ socially shared regulation. Furthermore, content analysis reveals that the frequency of socio-emotional behaviors exhibit differences across the various collaboration stages. This study confirms that emotion scaffolds effectively maintain a positive collaborative atmosphere, reduce task uncertainty, and improve collaboration efficiency. These findings provide a theoretical and practical foundation for designing emotional interventions in future CSCL environments.

ABSTRACT. Large language models often exhibit uneven performance across different knowledge domains, where a single general-purpose model may not perform equally well across all tasks. This study proposes an efficient expert-routing framework that dynamically directs multiple-choice questions to domain-specific expert models. The framework employs a routing module to classify questions into math, programming, or philosophy domains before forwarding them to specialized expert models for inference. Experiments were conducted on 300 samples extracted from the MMLU benchmark, with 100 questions evaluated per domain. The proposed framework achieved a domain routing accuracy of 99.0% and improved overall multiple-choice question answering accuracy from 52.3% to 63.3%, yielding an absolute improvement of 11.0 percentage points over the baseline model. The largest gain was observed in programming tasks, where performance increased by 18.0 percentage points and achieved statistical significance (p=0.0019). The overall improvement across all domains was also statistically significant (p=0.0013). These findings demonstrate that domain-aware routing can effectively enhance reasoning performance by leveraging specialized expert models instead of relying solely on a single general-purpose model.

ABSTRACT. This study addresses the gap between factual progress and cognitive perception arising when students fail to meet learning goals. We developed an interactive reflection support system incorporating the column method used in cognitive behavioral therapy (CBT) to foster adaptive thinking. In all, 10 university students participated in a 6-week crossover evaluation. The results indicated that the CBT-based chatbot structured reflections compared to the baseline. In particular, the intervention significantly improved the quality of introspective writing, shifting focus from mere description of the cause to the formulation of concrete, actionable strategies and adaptive reasoning. By addressing psychological barriers and cognitive biases, the system allowed students to translate negative emotions into proactive planning. These findings indicate that the integration of psychological frameworks with generative AI can enhance self-regulated learning, promoting deeper self-insight and a more resilient behavioral adaptation in academic research environments.

ABSTRACT. Given the educational efficacy of error correction, this work focuses on designing theory-guided error-correction learning activities to facilitate the attainment of significant educational goals. Noting the significance of social-emotional competence for long-term personal development, and in view of social awareness theory, a two-stage learning activity was proposed—correcting one’s own wrong answers in the first phase and correcting the class-wide most commonly incorrectly answered questions in the second phase. To examine its educational potential, one fifth-grade class (n=22) participated. Several major findings were obtained. First, results from one-sample t-tests indicated that perceived learning usefulness ratings were significantly higher than the scale midpoint value of 3 for both phases. Second, a paired-samples t-test found that the difference between the two phases was not statistically significant. Third, qualitative data indicated that students appreciated the two-stage error-correction learning activity for different reasons, highlighting their complementary functions: the first phase supporting personal error diagnosis, and the second phase supporting awareness of commonly shared misconceptions and difficult problem types. Overall, the lack of a statistically significant difference between the two phases, which were associated with distinct complementary functions, suggests that our theory-guided two-stage learning activity design has promising educational potential as an enriched error-correction activity that can contribute meaningfully to students’ learning.

ABSTRACT. In higher education, students often need accurate access to academic regulations, training policies, and administrative procedures. However, this information is usually organized in hierarchical structures, contains complex relationships, and is difficult to retrieve effectively using traditional text retrieval systems. This study proposes a student support chatbot based on Graph Retrieval Augmented Generation to improve the retrieval and interpretation of academic information in universities. The system constructs a knowledge graph from academic regulation documents, in which concepts and relations are represented as subject, relation, and object triples. Based on this structure, the system combines knowledge graph-based retrieval with semantic retrieval using a vector database. This approach allows the system to use both textual content and the relationships among regulation units, thereby providing more reliable context for the large language model during answer generation. Experiments on a set of questions related to student regulations show that the proposed system improves information retrieval compared with the baseline language model. Specifically, the system achieves a 21.20% improvement in Recall and outperforms the baseline in Recall in 74.60% of the cases. These results demonstrate the potential of Graph Retrieval Augmented Generation for building learning support and academic advising systems that can provide accurate, grounded, and student relevant information in higher education.

ABSTRACT. This study applies BERTopic, a neural topic modeling framework, to 51 empirical studies on creative, arts-based, and AI-enhanced pedagogies in health professional education, grounded in a sociomaterial theoretical framework. Seven coherent topics (49/51 documents) emerge across three axes: empathy development (n=12, 24.5%), clinical skills training (n=21, 42.9%), and methodological framing (n=16, 32.7%). Cross-method convergent validity is established through correspondence with deductive meta-analytic outcome domains. Critically, BERTopic surfaces a structural knowledge gap: equity, decolonial, and AI ethics vocabulary is absent across all topic outputs, constituting a sociomaterial condition in which certain communities and epistemologies remain unenrolled in the field's knowledge production infrastructure.

ABSTRACT. This paper proposes a machine learning-based adaptive self-assessment method for intersubjective evaluation using checklists and rubrics. The proposed method aims to reduce learners’ response burden while maintaining evaluation performance by sequentially imputing unanswered checklist items based on prediction confidence and subsequently estimating the rubric-based overall evaluation. Simulation experiments were conducted using university self-assessment data on eight soft skills collected between 2020 and 2025 under two settings: Single Self-Assessment (SSA), which uses only current responses, and Dual Self-Assessment (DSA), which additionally incorporates historical self-assessment data. The results demonstrated that the proposed method reduced the number of checklist items while maintaining acceptable evaluation performance across most skills. In particular, the DSA setting achieved higher reduction rates than the SSA setting, suggesting that historical self-assessment data can support more efficient adaptive self-assessment with fewer learner responses. These findings indicate that the proposed method has the potential to support sustainable and scalable intersubjective self-assessment practices in educational settings.

ABSTRACT. Self-regulated learning (SRL) is critical for translation performance, yet traditional training struggles with individual and contextual factors. While generative AI (GenAI) offers personalized feedback potential, concerns include AI hallucination, unknown training data sources, and uncritical student acceptance. AI-supported SRL has focused on reading and writing, leaving translation underexplored. This study examined how translation students' SRL skills and engagement changed using AI-Corpora, a platform integrating GenAI with corpora. Findings revealed different trajectories: forethought improved substantially early. Performance and reflection showed delayed but significant gains. Three dimensions of engagement progressed steadily. The study highlights GenAI-corpora integration's potential for fostering SRL and engagement while underscoring the need to support critical thinking and prevent over-reliance.

ABSTRACT. Inclusion in technology-enhanced learning is often framed in terms of access, accessibility, or the provision of digital tools. These concerns remain essential, but they are insufficient in learning environments shaped by platforms, analytics, adaptive systems, and artificial intelligence. Learners may be able to enter a digital system while still being unable to participate meaningfully, benefit fairly, or challenge data-driven decisions that affect their learning. This paper develops a conceptual framework for analyzing inclusion in intelligent technology-mediated learning ecologies. Drawing on a conceptual review informed by critical interpretive synthesis, it synthesizes research on accessibility, Universal Design for Learning, digital equity, online and hybrid learning, AI in education, and educational data governance. The review identifies four shifts: from accessibility compliance to inclusive design, from technical access to meaningful participation, from technology provision to digital equity and fair outcomes, and from human-centered design to participatory socio-technical governance. Building on these shifts, the paper proposes the Access-Participation-Equity-Agency (APEA) framework. APEA conceptualizes inclusion across four connected ecological layers: material conditions, pedagogical interaction, social distribution, and governance. An illustrative case of generative AI writing tools shows how the same technology can expand access and participation while also creating new equity and agency risks. The paper argues that inclusion in AI-mediated learning should not be reduced to access, usability, or personalization alone. It must also involve meaningful participation, fair distribution of benefits and risks, and the capacity of learners and educators to understand, influence, refuse, adapt to, or contest technology-mediated arrangements. The framework is particularly relevant to Asia-Pacific and ICCE-related contexts, where multilingual learning, uneven infrastructure, and differences in institutional capacity shape the adoption of intelligent technologies.

ABSTRACT. The premise of this study was to examine how the use of Information and Communications Technology (ICT) in inquiry-based learning (IBL) predicts academic achievement in mathematics, reading, and science, both directly and when mediated through student curiosity. To do this, we analysed the data from Finland and South Korea in the Programme for International Student Assessment (PISA) 2022. In the present study, ICT use in IBL consists of three different components: 1) Standard activities, 2) Investigative activities, and 3) Project management. Moreover, student gender and the index of economic, social and cultural status (ESCS) were included as covariates to consider possible confounding effects. Our results show that ICT use in standard IBL activities is a consistently positive indicator of academic proficiency in all subject domains in Finland and South Korea. Moreover, the predicted total effect was even greater in all three subject domains in Finland when mediated through student curiosity. ICT use in investigative activities, however, predicted a wholly negative effect on student academic achievement and in Finland, using ICT in investigative activities reduced student curiosity and further decreased the predicted academic proficiency in all three subjects. Our findings indicate that the way that ICT is used in IBL can influence both student curiosity and academic proficiency in PISA 2022.

ABSTRACT. This study investigates linking national digital skill standards (DSS-P) to knowledge maps of IT subjects, examining whether such links help learners develop career visions and understand learning steps. We developed a prototype system in which selecting a career role displays linked knowledge map learning items and their importance levels. A preliminary evaluation with university students (n = 9) suggested usefulness for career vision development, although more specific learning support is needed.

ABSTRACT. Our prior work proposed a four-dimension framework for the qualitative analysis of programming-video annotations. This paper concerns the first dimension, Note Type, which classifies annotations by surface form and has since served as an upstream classifier for cognition-aware feedback, dashboards, and deviation detection, yet has never been independently validated. We test its reproducibility on 1,471 annotations from three years of an undergraduate Python course, isolating the 87 system-tagged Questions and comparing six perspectives on the remaining 1,384 notes. We compare expert annotation, a rule-based codebook, unsupervised clustering with multilingual sentence embeddings and character n-grams, a supervised classifier trained by five-fold cross-validation on expert labels, and large language models under zero-shot and few-shot prompting conditions. The LLM conditions are run with both Claude Sonnet 4.5 and GPT-4o. Expert agreement reaches Cohen's κ = 0.83. The rule-based codebook reaches κ = 0.68 against expert consensus. Unsupervised clustering and LLM induction do not recover the taxonomy, instead defaulting to Python-topic structure rather than surface form. With expert calibration, two operationally distinct methods reach expert-comparable agreement: a supervised classifier over the codebook's own signals reaches κ = 0.78, statistically above the codebook, while a few-shot Claude classifier with 30 expert examples reaches κ = 0.73, not statistically distinguishable from the codebook. The supervised classifier and the few-shot LLM are also not statistically distinguishable from each other. These results show that the Note Type dimension is reproducible by calibrated automated methods at different expert-label budgets, but that this dimension is imposed by the codebook and expert examples rather than emerging naturally from the corpus. Whether the result extends to the framework's three deeper dimensions remains open.

ABSTRACT. Large language models are increasingly explored as interactive tutors for mathematics learning, making it important to understand how model adaptation and Socratic scaffolding each contribute to tutoring quality. This study reports a controlled quality–cost comparison of four Socratic AI math tutor conditions across 241 held-out algebra and geometry problems. The system prompt, student simulator, dialogue protocol, and evaluation framework were held constant. Composite scores were closely clustered: the supervised fine-tuned model achieved the highest score (0.634), but its gain over the prompt-only Llama baseline was only 0.013, and no planned contrast reached practical significance. In contrast, removing the Socratic scaffold reduced the composite score by 0.188. Cost analysis indicated that model adaptation and frontier API deployment did not provide clear quality–cost advantages under the study assumptions. These results suggest that, in simulated tutoring settings and under the tested prompt design, scaffold design may be a more immediately impactful design lever than model adaptation.

ABSTRACT. As Large Language Models (LLMs) are increasingly adopted for automated grading and feedback in higher education, their structured outputs, including multi-dimensional rubric scores, detailed feedback comments, and improvement suggestions, create an appearance of thorough analytic evaluation. This study examines whether these outputs deliver what they appear to offer. Using the JorGPT dataset of 3,041 student responses to 50 open-ended computer science questions, scored by both human instructors and three commercial LLMs, we identify three systematic discrepancies between the apparent and actual quality of LLM-generated grading and feedback. The sub-dimension scores are highly correlated (r = 0.82–0.99, VIF up to 45), providing redundant rather than independent diagnostic information. The textual feedback rarely detects student misconceptions (5–7% vs. 15–31% for teachers), functioning as a coverage checklist rather than a diagnostic instrument. The feedback tone remains uniformly positive regardless of response quality, lacking the severity modulation observed in human feedback. Additionally, grading accuracy varies significantly by knowledge domain, with procedural topics most reliable. These findings provide empirically grounded guidance on which aspects of LLM-generated grading and feedback can be relied upon and which require continued human oversight.

ABSTRACT. Intelligent tutoring systems increasingly use LLMs to classify dialogue moves and inform adaptive scaffolding decisions. However, aggregate accuracy may hide errors on dialogue turns that are important for pedagogical decision-making. We propose VALID-ITS, an audit framework that evaluates LLM dialogue coders across four dimensions: subgroup and metric audit, rewrite stability audit, context and boundary audit, and scaffolding risk audit. We evaluated LLMs and supervised classifiers across three public corpora: TalkMoves, CIMA, and MathDial. The results show that aggregate accuracy and macro-F1 can diverge sharply, errors can concentrate at speaker-turn boundaries, and some pedagogically important error pairs can produce high wrong-scaffold rates. In TalkMoves, prompt for reasoning to revoicing reached a WSR of 1.000 for one coder variant. These findings suggest that LLM dialogue coders should not be deployed in ITS based only on aggregate accuracy.

ABSTRACT. In creative activities, it is crucial for learners to actively regulate the cyclical process of SRL (forethought, performance, and self-reflection). This study treats daily trial-and-error in research activities as "learning materials" and proposes a support system to foster better self-regulated learners. Specifically, we developed a framework that structures SRL in research into three layers: (1) problem-solving activities (RA layer), (2) SRL activities (SRLA layer) for process control, and (3) meta-learning activities (MLA layer) for acquiring metacognitive knowledge from the process. The proposed system features functions to organize thought products into a chain of "questions" and "answers" (RA layer), externalize solution plans as a hierarchical structure of "goals and means" (SRLA layer), and structurally support activities to reconstruct "lesson learnt" (metacognitive knowledge) based on these processes (MLA layer). A practical implementation involving 14 students confirmed that the three-layered processes can be continuously executed and externalized within daily research activities. The results suggested that chronological activity logs functioned as "learning resources" that drive dialogue with supervisors, facilitating the refinement of specific metacognitive knowledge tailored to individual cognitive characteristics.

ABSTRACT. Generative AI (GenAI) is reshaping project-based and learning-by-doing courses by offering rapid procedural assistance, yet unrestricted support may encourage answer offloading and weaken teachers’ ability to orchestrate synchronous classroom learning. This study proposes a Hierarchical Digital Twin Ecology, a teacher-governed multi-agent architecture in which student-facing digital mentors provide step-based guidance, request evidence before escalating assistance, and aggregate real-time learning states to a teacher dashboard for timely classroom intervention. Its key mechanism is a state-aggregation and threshold coordination protocol that turns stalled progress, missing evidence, repeated failures, and help-seeking patterns into visible indicators for teacher noticing and synchronous intervention. Rather than advancing a stand-alone psychological theory, the study operationalizes identity-based motivation, the zone of proximal development, and the Mantle of the Expert as proximal identity scaffolding for evidence-first recovery. A 14-week quasi-experimental study compared three conditions in a robotics and Artificial Intelligence of Things (AIoT) project course with 120 students: LMS-only support, Butler-style execution support, and the proposed Digital Twin Ecology. Because intact classes were used, baseline project performance was examined using one-way ANOVA, and outcome comparisons used ANCOVA with prior project performance as the primary covariate. For RQ1, the ecology group showed the strongest evidence-first recovery profile, including the highest 48-hour recovery after setbacks (72.86%), evidence checkpoint completion (85.61%), evidence submission before help requests (69.41%), reattempts after setbacks, and debugging-note completion. For RQ2, the ecology group achieved the highest project performance (M = 87.45), especially in debugging quality and documentation/evidence explanation, indicating stronger diagnostic and evidential competence. For RQ3, qualitative records and student reflections suggested that recovery was supported by evidence-first help seeking, proximal role framing, and dashboard-triggered teacher intervention when shared bottlenecks became visible. By linking evidence checkpoints with teacher-facing analytics, the ecology translated individual recovery traces into actionable classroom coordination. Together, these findings suggest that teacher-governed GenAI can support individualized assistance while preserving students’ epistemic agency and strengthening classroom-level orchestration. The study contributes a practical design account of teacher–AI complementarity in project learning and offers a transferable coordination logic for AI-supported project classrooms, while acknowledging the limits of its single-course, intact-class design and context-specific 48-hour recovery indicator.

ABSTRACT. Whether AI tools support genuine self-regulation—or merely simulate it—depends on how human-AI interaction is structured. We present ScholarMate, a four-agent LLM reading assistant instantiating a co-regulation loop: Navigator (system-initiated external regulation) works in concert with three learner-activated agents—Lexicon, Dialectician, and Reviewer—whose @mention selection requires learners to diagnose their own difficulty type. Deployed over ten weeks in a graduate course at Beijing Normal University (N = 35, 12,442 interaction records), ScholarMate generated two structurally distinct external regulation signals: section-transition triggers (metacognitive challenge questions at chapter boundaries) and idle-reminder triggers (redirection after extended dwell time). Lag sequential analysis revealed qualitatively different co-regulation effects: section triggers produced high learner activation (78.1% response rate, Agent-selection profile matching the overall voluntary baseline), while idle-reminder triggers produced lower activation (33.8%) but a significant directional shift toward Dialectician (+6.2 pp over baseline), suggesting learners infer deeper processing difficulty from extended dwell time. K-means clustering (k = 3) identified three co-regulation profiles—Low-Engaged (n = 8), Deep-Engaged (n = 14), and Surface-Active (n = 13). Message-length asymmetry confirmed qualitatively distinct self-regulatory acts across Agents (Lexicon M = 795 vs. Dialectician M = 71 characters; H = 925.51, p < .001). Reviewer's persistent underuse (M = 4.8) identifies self-reflective regulation as the primary co-regulation gap. Four design principles for authentic co-regulation loops in LLM-based learning environments are proposed.

ABSTRACT. Automated Short Answer Grading (ASAG) leveraging Large Language Models (LLMs) offers significant potential for reducing teacher workload in educational assessment. However, applying these systems to morphologically complex, low-resource languages like Thai presents unique challenges. A critical but underexplored limitation is "keyword dependency" the tendency of Artificial Intelligence (AI) to evaluate responses based on surface-level lexical matching rather than genuine semantic understanding, creating a grading inequity where students who paraphrase correct answers may be penalized, while those who use expected vocabulary without understanding may receive inflated scores. This study investigates the robustness of five major LLMs, namely ChatGPT-4o, Gemini 2.0 Pro, Claude 4.0 Sonnet, Grok 4, and Typhoon 2 Instruct (70B), against adversarial student response conditions in grading open-ended Thai-language written exams. As a core methodological contribution, we introduce a controlled adversarial testing design using six behavioral profiles: four representing normal student performance levels, and two adversarial cases, a Keyword Adversarial profile (responses containing expected keywords but with intentionally distorted semantics) and a Semantic Adversarial profile (semantically correct responses that deliberately avoid expected rubric keywords). All models were evaluated against a baseline of two professional Thai teachers using a 55-question dataset, assessed through a comprehensive 5-metric evaluation framework covering both numerical score agreement and explanatory faithfulness via reason codes. Under normal student response conditions (Students 1–4), the LLMs achieved MAE ranging from 0.218 to 0.400. Under the Keyword Adversarial condition (Student 5), the models exhibited clear keyword inflation, overscoring semantically flawed answers simply because they contained expected vocabulary. Most critically, under the Semantic Adversarial condition (Student 6), the models demonstrated extreme semantic undervaluation, systematically penalizing correct answers that utilized paraphrasing or synonyms, with Weighted Cohen's Kappa collapsing to as low as 0.067 and Krippendorff's Alpha falling well below acceptable reliability thresholds at 0.422. Reason code analysis directly confirmed the AI's structural reliance on lexical cues over logical correctness, revealing that current AI systems penalize keyword absence more severely than semantic incorrectness, a finding with direct implications for grading fairness in synonym-rich languages like Thai.

ABSTRACT. Generative artificial intelligence (GenAI) has been increasingly used as a learning tool in higher education. However, students’ final answers or quiz scores may not fully show how they engage with course content. This study examined how structured and unstructured GenAI-supported reflection made students’ learning performance, cognitive engagement, and self-review processes visible in an undergraduate course. A total of 28 students participated in a four-week learning activity using TronClass with GenAI-supported functions. The control group completed open-ended reflection tasks with GenAI support, whereas the experimental group completed a Bloom-based structured worksheet and a GenAI self-review form. Data included weekly AI-generated quiz scores, worksheet responses, and GenAI self-review responses. Worksheet responses were coded according to Bloom’s cognitive levels and grouped into surface-, middle-, and deep level cognition. Self-review responses were coded for conceptual awareness, error awareness, self-monitoring, self-regulation, and transfer of learning. The results showed that the experimental group had higher weekly quiz scores across the four weeks. Both groups produced evidence of surface-, middle-, and deep level cognition, but the experimental group showed more consistent distributions across cognitive levels. The GenAI self-review responses further showed students’ awareness of concepts, errors, unclear knowledge, learning strategy changes, and transfer to real-world scenarios. These findings suggest that Bloom-based structured worksheet and GenAI self-review forms can provide teachers with observable information about students’ learning processes when GenAI is used as a learning tool.

ABSTRACT. Primary and secondary online learners typically support to fully engage in learning activities. The Academic Communities of Engagement (ACE) framework was developed to explain how learner engagement is influenced by support provided by actors within students’ course and personal communities. However, the framework predates the widespread adoption of generative artificial intelligence (GenAI), creating a need to reconsider how emerging AI technologies may influence learner engagement and support. This presentation will revisit the ACE framework in light of recent developments in GenAI. First, the paper reviews the ACE framework’s three dimensions of engagement and the support indicators associated with each dimension. Next, it reviews emerging literature on the educational uses of GenAI and examines how AI-generated support can complement, extend, or potentially undermine human-provided support. It argues that GenAI should not be viewed as a replacement for teachers, parents, peers, or other support actors, but rather as an additional layer of support that must be guided by human oversight. The result is an updated ACE framework that integrates GenAI as a support mechanism within online learning environments and provides researchers and practitioners with a conceptual foundation for better understanding support systems that leverage both human and GenAI support to enhance learner engagement.

ABSTRACT. Generative AI is increasingly used to create educational content, yet most tools continue to position instructors as users of preconfigured systems rather than authors of learning environments. This paper examines how instructors engage in AI-mediated authoring of interactive learning simulators through vibe coding—a conversational, intent-driven practice in which pedagogical goals are translated into executable simulation logic through natural-language interaction, without formal programming. Focusing on a teacher-authored market sizing simulator developed and deployed in undergraduate management courses, we analyze authoring artifacts, parameter structures, and iterative design traces to identify how instructors formalize pedagogical judgment through model constraints, variable relationships, and scenario logic. Three design insights emerge: (1) pedagogical intent is expressed primarily through constraint-setting rather than content specification; (2) vibe coding functions as a reflective practice that externalizes tacit teaching judgment; and (3) instructor authoring redistributes teaching labor upstream into pre-class design, reshaping classroom orchestration. We argue that vibe coding represents an emergent form of AI-mediated instructional design that lowers technical barriers while preserving instructor agency over the structure of learning environments. The paper contributes design considerations for faculty-facing AI authoring tools and situates vibe coding within broader discussions of human–AI collaboration in educational technology.

ABSTRACT. Generative artificial intelligence (GAI) is increasingly reshaping creative learning practices, yet limited research has examined how pre-service teachers (PSTs) experience and adopt GAI in multimodal pedagogical design. This qualitative study investigates PSTs’ perceptions of GAI-supported digital storytelling (DST) through the lens of Rogers’ Diffusion of Innovation theory. Seventy-nine PSTs participated in an 18-week DST project in which they used tools such as ChatGPT, Midjourney, Runway, and CapCut to develop scripts, storyboards, images, animations, and final digital stories. Data were collected from reflective journals, focus group interviews, and stimulated recall interviews, and were analyzed thematically according to five perceived innovation attributes: relative advantage, compatibility, complexity, trialability, and observability. The findings show that PSTs perceived GAI as a productive co-creation partner that accelerated script development, visual generation, animation prototyping, and collaborative decision-making. At the same time, they encountered challenges related to prompt formulation, output inconsistency, cross-tool coordination, technical instability, and the pedagogical appropriateness of GAI-generated content. Participants responded to these challenges by refining prompts, combining multiple tools, post-editing outputs, negotiating design choices within groups, and iteratively aligning GAI-generated materials with their narrative and educational goals. The study suggests that GAI adoption in teacher education is not a linear acceptance process but a situated process of human-GAI negotiation, creative adaptation, and pedagogical judgment. These findings contribute to understanding how intelligent technologies can be integrated into teacher education to support creative, collaborative, and reflective learning ecologies.

ABSTRACT. In organizational knowledge creation, it is important to share and accumulate the insights and judgments that members gain through discussions and to develop them into reusable knowledge within the organization. However, in discussions related to research activities, although outcomes and conclusions are often shared, the tacit understandings underlying them, such as assumptions and criteria for judgment, are not always made explicit. In this study, we developed a support system that records utterances in which tacit knowledge is perceived to have been activated during discussions, formalizes them as Knowledge Fragments after the discussion, and further supports their construction and accumulation as organizational knowledge. An initial practice in a research laboratory suggested that the system has the potential to support activities in which individual insights are developed into organizational knowledge.

ABSTRACT. Modern industrial education needs new digital tools to teach practical skills. While Information and Communication Technology (ICT) is now widely used in regular classes, vocational high schools and technical colleges still face a serious problem. Skilled engineers use many senses—sight, hearing, touch, and the feeling of force—to judge their work. This "sensory tacit knowledge" is very hard to teach through videos or words alone. This study proposes a new e-learning material, called "E-learning by Doing," that uses a sensory transmission device to transfer this knowledge to learners. As a first step, we focused on the manual feed operation of a lathe, where keeping a stable feed rate determines surface quality. The proposed material delivers an expert's feed-rate motion to learners as direct physical stimuli, helping them calibrate their own speed sense without conscious imitation. To evaluate the material, 27 beginners participated in an experiment as a control group (12, video only) or an experimental group (15, video plus device), and performed face and taper machining on a real lathe. In face machining, the experimental group showed improved surface profile quality, especially in Pa (ANCOVA: Pa p = 0.025, η² = 0.192, large effect). The correlation analysis showed that the control group followed a single stabilization path (ρ=-0.713, p=0.009), whereas the experimental group achieved good quality through different motion strategies.

ABSTRACT. Component-based reconstructive learning has already been used in several domains as a practical design in which learners reconstruct a meaningful structure from given components. In this paper, reconstruction does not mean simply assembling arbitrary components. Rather, it refers to an activity in which an intended reference structure is decomposed into components, and learners reconstruct it from a component set designed to be necessary and sufficient for reconstructing that structure within the task. This constrained component set reduces the burden of generating components from scratch while requiring learners to interpret component meanings, judge semantic relations, and organize them into a coherent structure. However, because components are presented in advance, this approach has remained insufficiently theorized as a form of constructive learning. This paper provides a theoretical foundation for positioning component-based reconstruction as a legitimate constructive learning design by distinguishing free externalization from reconstructive externalization. Free externalization involves expressing already segmented understanding, forming and adjusting understanding through externalization, and leaving some target content unexternalized. Reconstructive externalization, by contrast, externally supports part of component generation and focuses learners’ activity on bridging the semantic gap between components and structure. By organizing this relation through the concepts of the duality of externalization, cyclic difficulty, semantic gap, and unreconstructed content, this paper clarifies why reconstruction-based learning should be recognized not as a simplified substitute for constructive learning, but as an educationally useful design that supports learners’ participation in structure construction and strengthens its theoretical intelligibility and adoptability. Keywords: Reconstruction-based learning, Concept map, Semantic gap

ABSTRACT. On-demand video learning has become increasingly popular due to its flexibility, accessibility, and support for self-paced learning. In these on-demand video- based learning environments, students often struggle to maintain their attention due to the lack of real-time learning support. To address this issue, this paper proposes a robot-mediated learning support system that integrates real-time state estimation and adaptive intervention. The system continuously monitors learners’ attention states using multimodal signals, including gaze tracking and head-pose estimation. Based on these signals, the system identifies learners’ cognitive states. When distraction is detected, a physical social robot dynamically provides prompts to help learners refocus their attention. A quasi-experiment was conducted by recruiting 20 participants. The result showed that the robot-assisted group approach effectively improved participants’ learning achievement. Temporal analysis based on the Eye Aspect Ratio (EAR) further demonstrated greater stability of attention, and attentional recovery also improved following the intervention. Questionnaire results indicated improved concentration (M = 3.69 vs. 3.25), although cognitive load slightly increased (M = 3.06 vs. 2.87). These findings suggest that real-time, context-aware robot intervention effectively supports sustained attention while highlighting a trade-off between attentional support and cognitive burden.

ABSTRACT. In a tertiary-level Research Methods in Computing course, students perform cognitive tasks, primarily literature review and synthesis, to formulate research questions (RQs) that are feasible and aligned with identified gaps. The iterative development of these RQs necessitates continuous pedagogical guidance to maintain alignment with the eventual research scope and design. The fluent linguistic and reasoning abilities of Large Language Models (LLMs) make them prime candidates to assist students in this process. This paper describes a study in designing learning activities that explicitly instruct students to employ Generative AI (GenAI) in refining their RQs and overall proposal document while mandating a formal declaration of AI usage. Findings from a mixed methods survey and analysis of student artifacts reveal critical gaps in AI literacy, specifically regarding how students interpret AI feedback and integrate it in refining their work. While students utilized GenAI for various research tasks, their self-assessments indicated disparity in AI usage collaborative practices, perceived reliance, and awareness of academic policies. Findings from this study suggest a two-pronged approach to AI literacy: (1) empowering educators to leverage LLMs for immediate, scalable feedback on student work, and (2) training learners to apply critical thinking when evaluating AI-generated output. The paper concludes with pedagogical insights and recommendations for educators seeking to position LLMs as a pedagogical agent, specifically critic and collaborator, rather than a mere writing tool within the research process.

ABSTRACT. Coding of learner behavior is a basic methodology in educational research, but its automation in physical-space settings still relies heavily on manual annotation, which constitutes a major bottleneck for scaling up video-based analysis. Classical computer vision and machine learning approaches typically require per-study dataset construction and fine-tuning, which is difficult given that educational data is not always captured in the same environment and due to limited model flexibility. In this study, we propose a method for automatically estimating the object on which a learner is fixating. The method takes as input a first-person video recorded by a wearable eye tracker, with the learner's gaze position overlaid on each frame, and uses a Vision-Language Model (VLM) guided by natural-language class definitions to classify the fixated object. Overlaying the gaze position bypasses gaze-direction inference, at which VLMs underperform, while leveraging their strength in open-vocabulary recognition. We applied the method to a Russian-as-a-second-language reading and writing task performed by five university learners, classifying the fixated object at 10-second intervals into five researcher-defined classes. Compared with human-annotated ground truth, the VLM achieved an accuracy of 0.870 and a Cohen's kappa of 0.781. The results suggest that the proposed method can classify the objects indicated by overlaid gaze markers with substantial agreement with human annotation.

ABSTRACT. In inquiry-based learning, it is crucial for learners to examine inquiry topics　divergently from multiple perspectives, formulate consistent and logically coherent exploratory hypotheses aligned with their objectives, and use these hypotheses to guide subsequent exploration. However, when learners’ focus on exploratory keywords becomes narrow, the range of hypotheses available for critical examination is correspondingly limited. To address this issue, we developed a support system equipped with two functions: one that presents candidate keywords to help learners expand and organize their keyword maps, and another that provides advisory prompts based on the SCAMPER method, a well-known idea-generation technique, to encourage the revision and structuring of exploratory hypotheses. An evaluation experiment involving 12 students showed that the numbers of both keywords and hypotheses increased after the introduction of these support functions, indicating the potential of the system to facilitate the formulation of multifaceted exploratory hypotheses. Furthermore, questionnaire results indicated that participants generally perceived the support for organizing keywords and formulating hypotheses as useful.

ABSTRACT. In classroom lectures, the instructor's body might block blackboard content. This occlusion degrades learning in cognitive load theory and downstream AI tools that process lecture videos. A survey about AI educational systems reveals that none of them involves occlusion handling. To bridge this gap, we present an innovative approach named Borrat to detect and remove teacher occlusions and restore blackboard content using computer vision techniques. On a real-world classroom dataset with teacher occlusions, our model achieves 97.99% performance, with only 4.5% latency overhead. We also evaluated our technique on two real-world MIT linear algebra lectures, and restored content regions yielding 8.9% more OCR characters. Moreover, we generated flashcards and multiple-choice questions from extracted text and classified them using Bloom's Revised Taxonomy. Results show that restored videos can produce more high-quality questions in higher-order thinking. We discuss the temporal split-attention problem in education based on cognitive load theory and how our technique addresses it, as well as its implications and applications in concrete use cases.

ABSTRACT. Designing well-calibrated assessments in large-scale online degree programmes involves a structured, multi-stage review process in which instructors explicitly decide the question type, mark allocation, and cognitive difficulty level of each item. We argue that this institutional process encodes item difficulty and discrimination, as formalised by the Two-Parameter Logistic (2PL) Item Response Theory (IRT) model, into the observable structural features of the questions, and we test this hypothesis empirically using eleven consecutive terms of a Statistics course from an online degree programme at a tier-1 Indian university (181 questions, 4,649 students). Ground-truth item parameters are estimated via an Expectation-Maximisation(EM) algorithm applied to historical student response data. We evaluate five regression models across all 64 subsets of six-question feature groups (320 combinations) under a strict longitudinal expanding-window protocol. The principal finding confirms the hypothesis: question type and mark value are the most informative features for predicting both difficulty and discrimination. A simple conditional mean estimator conditioned solely on these two features achieves a mean combined Root Mean Squared Error (RMSE) of 0.9784, closely approaching the best machine learning estimator (Random Forest (RF), 0.9126), and requires no model fitting beyond a lookup table of historical averages. This has a serious implication for assessment designers as it provides a more explainable and robust mental model while designing assessments for this course in future. As a downstream application, predicted item parameters are combined with prerequisite-course-based student ability estimates via the 2PL response function to produce student quiz score predictions before any current-term response is collected, achieving a score RMSE of 33.755 against an oracle lower bound of 28.350. These results provide both a practically deployable tool for pre-assessment calibration and a data-driven explanation of why structural design choices reliably encode item psychometric properties.

ABSTRACT. This study investigates an instructional approach integrating GenAI, embodied cognition (EC), and argumentation-based instruction (ABI) to support junior high school students’ AI learning. Thirty-eight students participated in a two-day program involving conceptual instruction, embodied activities, GenAI tasks, and the AI2 Robot City board game. Results showed significant improvements in students’ learning effectiveness and metacognitive awareness. Behavioral sequence analysis further revealed connections between embodied actions and argumentative discourse, suggesting that students engaged in reasoning, coordination, questioning, and collaborative knowledge construction. As an exploratory study, the findings provide initial evidence for integrating EC and ABI in GenAI-supported learning environments.

ABSTRACT. Alzheimer’s disease is a neurological disorder that deteriorates a person’s brain which impairs cognitive skills such as thinking, reasoning, and memory. Despite the Philippines’ aging and affected population, awareness on the disease is lackluster as misconceptions are still present in the media. This aim of the study is to bring more awareness regarding the effects of Alzheimer’s disease on the patient and their interpersonal relationships through a narrative-driven video game titled The Long Goodbye. The mechanics are based on research of the cognitive impacts of Alzheimer’s disease, while the story is influenced by conducted interviews with Filipino caregivers. As such, this study explores how game design can serve as a medium for education and emotional engagement regarding Alzheimer’s disease. From a round of playtesting, it was observed that the narration in the game helped deepen the understanding of the players on the social issue. However, there is still room for improvement in terms of specifying how the mechanics the players experience relate to the issue rather than merely just an aesthetic choice.

ABSTRACT. Online learning faces persistent challenges in sustaining student engagement, and while adaptive gamification has been proposed to address the inconsistent effects of one-size-fits-all gamification approaches, most implementations rely on static adaptation that overlooks the temporal evolution of learner motivation. This paper presents the design, implementation, and preliminary evaluation of a three-stage Dynamic Gamification Adaptation (DGA) algorithm. The algorithm establishes player types via the validated Hexad scale, maps each player type to a set of game elements grounded in the established learning theory, and monitors weekly engagement trajectories to trigger voluntary player type re-assessment when sustained motivational drift is detected across a three-week cycle. Implemented as a Moodle plugin and evaluated across three university courses involving 135 students, the system successfully differentiated all six Hexad player types and the re-assignment mechanism operated with reasonable selectivity, identifying only students exhibiting persistent engagement decline rather than transient fluctuations. These findings provide proof-of-concept support for treating learner motivation as a dynamic trajectory and suggest that adaptive gamification systems may benefit from combining structured player type differentiation with learner-initiated re-alignment mechanisms.

ABSTRACT. Despite the availability of sophisticated inquiry-based labs, literature reports persistent issues such as low learner engagement. This often stems from the daunting nature of complex problem scenarios, intangible scientific concepts, and cognitively demanding tasks inherent in inquiry-based learning environments. As a result, learners may feel overwhelmed and hesitate to explore underlying phenomena, struggle to complete the required cognitive activities, underutilize the affordances of associated simulations, or even drop out prematurely. These issues underscore the need for design interventions that can support and sustain cognitive engagement during inquiry-based learning. This paper presents the design of GaMINLab (Gamification (Meaningful) in INquiry Lab), that employs content gamification approach to cognitively engage learners in science inquiry practices in simulations-based labs. Unlike structural gamification, Content gamification integrates game-like elements directly into the learning content to enhance cognitive involvement. In GaMINLab, learners interact with problem scenarios built around OLabs simulations by engaging in core inquiry practices such as planning and carrying out investigations, analyzing and interpreting data, engaging in argument from evidence, etc. Relevant content gamification elements such as problems based on storyline, role based avatars, progression indicators, task-specific badges, etc and supporting pedagogical elements are incorporated in and around these inquiry tasks. We report student perceptions of these elements across 4 iterative studies where they engaged with inquiry tasks in GaMINLab. Qualitative data from student feedback and audio transcripts were analyzed using a deductive thematic analysis. We investigated impact perceived by student on affective domain overall, science process skills related, Scaffolding/support. Findings reveal that avatars and progression indicators’ novelty effect diminishes across multiple problem-solving cycles, while for elements like story-based problems, task specific badges the effect persists. The findings also highlight the role played by each element in isolation and together in the overall gamification ecosystem during the problem-solving process.