Guided by the Inquiry Cycle framework, this study addresses the challenge of providing individualized support in Virtual Inquiry-Based Learning (VIBL). We designed and evaluated the AI Curator, an LLM-based pedagogical agent that provides phase-specific support tailored to five inquiry phases: orientation, conceptualization, investigation, conclusion, and discussion. A quasi-experiment involving 69 fifth-graders in authentic science classrooms compared LLM-based support (n = 32) against a rule-based baseline (n = 37). ANCOVA results indicate that the LLM-based group achieved significantly higher post-test performance (F(1, 66) = 4.48, p = .038). Beyond outcomes, process analytics revealed that LLM-supported students spent significantly more time in the orientation and conceptualization phases while requiring fewer interactions during investigation and discussion. A key contribution lies in the characterization of a behavioral transition from procedural compliance to reflective knowledge construction. Lag sequential analysis (LSA) confirmed a significantly stronger reflective loop, transitioning from Revise Hypothesis to Experiment (Z-diff = 4.29, p < .001), in the LLM group. Discourse analysis further suggests these gains were driven by the AI Curator’s use of phase-sensitive tutoring moves, specifically cognitive restructuring and positive valorizing. These findings provide empirical evidence for designing LLM-based support to foster multi-phase inquiry in virtual environments.

Gamified project-based learning (GPBL) is an innovative pedagogy that combines the core principles of Project-Based Learning (PBL) with engaging games or gaming elements to enhance student motivation, participation, and learning outcomes. This study investigated the impact of GPBL compared with traditional PBL on students’ academic performance and learning motivation in elementary mathematics classes. Conducted over 2 weeks with 48 sixth-grade students in China, a quasi-experimental design was applied with an experimental group and a control group, each experiencing 10 learning sessions. Pre- and post-assessments measured students’ mathematics achievement and motivational shifts, revealing that GPBL enhanced both academic outcomes and motivational dimensions significantly more than PBL did. Qualitative analysis of interviews with students and teachers highlighted the role of gamified elements—such as role-playing, competitive challenges, and interactive video games—in fostering a more engaging and effective learning environment. The results offer practical insights for the integration of gamification in PBL frameworks to optimize instructional strategies and support student motivation and achievement. A theoretical framework for GPBL is also proposed based on the experimental results.

Bullying harms learning and wellbeing, and schools increasingly use digital games as part of their response. This PRISMA-guided systematic review and meta-analysis synthesized 34 studies on educational games for school bullying and cyberbullying intervention. Ten randomized trials passed risk-of-bias screening; nine contributed continuous outcomes to the primary meta-analysis. The pooled effect was small and statistically significant (Hedges' g = 0.14, 95% CI [0.02, 0.26]) and remained stable across sensitivity analyses. Effects were observed at immediate post-intervention assessment and tended to diminish by three to six months without booster components. Among outcome domains, emotional outcomes showed the largest pooled gains, while cognitive and behavioural effects were smaller and did not reach statistical significance. Educational games outperformed usual-care conditions, were comparable to other active programs, and showed higher completion rates than no-intervention controls. Implementation was predominantly classroom-based on PC/web platforms and targeted whole student cohorts rather than role- or risk-defined groups. Common design features included role-based scenarios, branching decisions, immediate feedback, and structured reflection. No tested moderator reached statistical significance. Future trials should pre-specify maintenance strategies, adopt role-sensitive recruitment, use well-matched active comparators, and report design and implementation features with greater precision.

Understanding how game design features shape learners' emotional dynamics is crucial for advancing digital game-based language learning from a human-computer interaction perspective. This study investigated the associations between interface color features, gameplay activities and emotional transitions in a 3D role-playing digital game-based language learning environment. Facial expression recognition was applied to continuously track affective states, with emotional transitions categorized into four outcomes: positive, negative, neutral, and engagement. Results indicated that transitions occurred most frequently under cool hues with low-to-medium saturation and high brightness, with cyan-green to blue-green tones (80-100) producing the highest transition ratios. Engagement and neutral transitions predominated in these conditions. Activity-level analyses further revealed distinct patterns: conversation-based tasks stimulated positive transitions, feedback and response activities elicited negative transitions, and large language model (LLM)-based tutor dialogues primarily guided transitions toward engagement and neutral states. These findings demonstrate that both interface design and in-game activities were systematically associated with emotional transitions in DGBLL, offering practical guidance for the development of emotionally responsive educational games.

Embodied agents, which utilize physical or virtual representations to simulate human-like social cues and multimodal interactions, have gained significant traction in educational settings. However, while empirical studies on embodied agents in education have grown significantly, there remains a lack of systematic reviews to synthesize their application status, technological evolution, and theoretical mechanisms, particularly concerning the paradigm shifts in “embodiment” and “inquiry” brought by large language models. Therefore, this study systematically reviews and analyzes 47 core empirical studies on embodied agents in education, detailing their developmental trajectories, functional characteristics, and outlining future research directions. Our findings indicate that embodied agent applications are primarily concentrated in domains with high cognitive load and strong empathy demands, such as healthcare, social sciences, and STEM. Technologically, these agents have undergone three paradigm shifts: from rule-driven pedagogical agents to adaptive learning companions, and currently to GenAI-driven conversational inquiry agents. Furthermore, the empirical evidence highlights that their intervention mechanisms are deeply rooted in cognitive load theory, social agency theory, and self-regulated learning theory, forming a multidimensional functional system that encompasses emotional engagement, cognitive and inquiry support, empathy cultivation, and social collaboration. Our review also elaborates on essential future directions, emphasizing the need to construct stage-specific cognitive scaffolds for complex inquiry, deepen metacognitive guidance for human-machine shared regulation, explore multi-agent collaborative ecosystems, and mitigate the implicit biases of GenAI. These insights provide valuable guidelines for future empirical research and the design of intelligent educational environments.

The cultivation of computational thinking is an urgent task, and programming education is an effective way to foster this skill. Given the current challenges in programming courses, such as high difficulty levels and low student engagement, game-based learning, particularly learning by making games, may offer new and meaningful solutions. This study summarizes the existing research on learning programming through making games, and discusses the importance of designing and developing a programming education platform oriented towards making games. Based on this, this study presents the design and development of the programming education platform MetaDream under the guidance of constructivism and further introduces its key educational features. MetaDream, as a programming education platform primarily aimed at game creation, allows students to freely build scenes and engage in dynamic programming interactions within the environment, thereby enhancing their computational thinking during the creation process.

Educational games, spanning a wide spectrum from digital games to non-digital games, have been considered as promising tools to facilitate math learning. However, previous studies focused more on digital game-based learning, few compared the differences between digital and non-digital games. We conducted a quasi-experiment to examine the effects of digital and non-digital games on math knowledge acquisition, intrinsic motivation and cognitive load in Chinese classrooms. Two equivalent versions of games (digital vs non-digital) focusing on coordinate knowledge acquisition were developed, and three groups (digital game, non-digital game and traditional teaching group) were assigned. The results reveal that students in the digital game group reported significantly higher intrinsic motivation and lower intrinsic and extraneous cognitive load than non-digital game group. While non-digital game group had better knowledge acquisition performance and germane cognitive load than digital game group. Our findings suggest that affordance of game technology influences students’ motivation and cognition, causing “false” high motivation and low cognitive engagement among students. Non-digital games could serve as cost-effective alternatives to enhance math learning performance within authentic classrooms, thus providing valuable insights into establishing a resilient and sustainable game-based learning environment.

Despite the potential of using ChatGPT to augment language learning, its impact on enhancing English as a Foreign Language (EFL) learners' reading comprehension and critical thinking is under-researched. This quasi-experimental study aimed to explore the effectiveness of integrating ChatGPT into EFL reading activities through a competitive questioning approach. In the five-week experiment, 87 Chinese university freshmen were divided into experimental and control groups to participate in weekly learning sessions in which they conducted reading exercises and question-posing activities. During the sessions, the experimental and control groups were respectively engaged in pursuing ChatGPT-supported competitive questioning and traditional text-based questioning. With the use of machine learning techniques, all questions posed by the participants were analyzed with a six-dimension framework of critical thinking, adapted from Facione's critical thinking framework and Day and Park's reading comprehension taxonomy, encompassing simple and compound interpretation, simple and complex analysis, synthesis, and inference. Results showed that the experimental group achieved significant gains in reading comprehension from pre-test to post-test, outperforming the control group, whose performance declined over the same period. Regarding critical thinking, the experimental group's posed questions were progressively more advanced in the course of the experiment. The present work underscores the potential of utilizing ChatGPT as a supportive competitor for EFL learners to improve reading comprehension and foster critical thinking.

Traditional teacher interpersonal communication training is often limited to face-to-face formats, offering few opportunities for practice and relying on fixed role-play scenarios. Artificial intelligence (AI) presents a solution to these limitations. In this study, we developed a ChatGPT-4.0-based platform for pre-service teachers to practice and enhance their interpersonal communication skills. A total of 29 pre-service teachers par-ticipated in a one-month training program on the platform, where they could repeatedly simulate real-world communication dilemmas and receive iterative feedback from generative AI. The results demonstrated that the training significantly improved the teachers' interpersonal communication skills and teaching efficacy. This study highlights the potential of AI to empower teacher professional development.

Spatial ability refers to the ability of individuals to recognize,encode,store,represent,decompose,combine and abstract objects or spatial figures in their minds,which is the cognitive foundation for understanding one's environment and solving problems.Building an accurate,convenient and effective assessment system of spatial ability is of great significance to the enhancement of STEM education and the quality of talent cultivation.Due to the complex,multi-dimensional and implicit nature of spatial ability,it is difficult to evaluate spatial ability via computer-based assessments.This study aims to accurately,effectively,and massively evaluate spatial ability by using multimodal learning analytics methods to explore the characteristic behavioral expressions of learners'spatial cognition,and by developing key technologies and tools for spatial ability stealth assessment based on video game environments.The specific contents include:1)Construct a framework for the intrinsic representation of spatial ability and an evaluation index system;2)Constructing a learner spatial ability behavior performance model based on multimodal learning analysis;3)Explore the key factors that influence spatial ability in video games,and use game engines to develop game-based assessment tools;4)Use evidence-centered design frameworks and Bayesian network models to develop and deploy assessment algorithms capable of inferring and predicting spatial abilities;5)Conduct empirical research in laboratory and real classroom settings to verify the effectiveness of evaluation tools.The research findings will contribute to a better understanding of human spatial cognition processes and behavioral performance,expand and enrich theories related to spatial abilities,and provide key technical support for large-scale digital assessment.

Help-seeking is an active learning strategy tied to self-regulated learning (SRL), where learners seek assistance when facing challenges. They may seek help from teachers, peers, intelligent tu- tor systems, and more recently, generative artificial intelligence (AI). However, there is limited empirical research on how learners’ help-seeking process differs between generative AI and hu- man experts. To address this, we conducted a lab experiment with 38 university students tasked with essay writing and revising. The students were randomly divided into two groups: one seek- ing help from ChatGPT (AI Group) and the other from an experienced teacher (HE Group). To examine their help-seeking processes, we used a combination of statistical testing and process mining methods, analyzing multimodal data (e.g., trace data, eye-tracking data, and conversa- tional data). Our results indicated that the AI Group exhibited a nonlinear help-seeking process, such as skipping evaluation, differing significantly from the linear model observed in the HE Group which also aligned with classic help-seeking theory. Detailed analysis revealed that the AI Group asked more operational questions, showing pragmatic help-seeking activities, whereas the HE Group was more proactive in evaluating and processing received feedback. We discussed factors such as social pressure, metacognitive off-loading, and over-reliance on AI in these different help-seeking scenarios. More importantly, this study offers innovative insights and evidence, based on multimodal data, to better understand and scaffold learners learning with generative AI.

The integration of Artificial Intelligence (AI) into K-12 education holds significant potential for teacher empowerment, yet existing research often neglects the heterogeneity of teachers and the diverse pathways through which empowerment unfolds. This study investigates how AI empowers teachers in China’s K-12 sector, responding to national AI initiatives and addressing the overlooked diversity of teacher experiences. Using a qualitative multi-case design, five teachers were purposefully selected to capture variation in teaching experience, technical proficiency, subject specialization, geography, and pedagogical philosophy. Data were gathered through interviews, classroom observations, and AI usage logs over four months. Analysis reveals that empowerment is not linear but emerges through interconnected processes of efficiency-driven adoption, pedagogical innovation, and ethical reflection. The proposed three-tiered model underscores the situated and differentiated nature of AI integration, shaped by individual characteristics and contextual factors. These findings offer practical implications for policymakers and school leaders, highlighting the need for differentiated support ranging from technical training to ethical guidance. By foregrounding teacher agency and contextual variation, the study advances international discourse on AI in education and emphasizes that the transformative potential of AI depends as much on teachers’ values and professional judgment as on technological affordances.

ChatGPT, which rides on generative artificial intelligence (AI) technology, has garnered exponential attention worldwide since its release in 2022. While there is an increasing number of studies probing into the potential benefits and challenges of ChatGPT in education, most of them confine the focus to informal learning and higher education. The formal adoption of ChatGPT in authentic classroom settings, especially in K-12 contexts, remains underexplored. In view of this gap, we conducted a quasi-experiment to study the effects of integrating ChatGPT into a compulsory Grade-10 English as a foreign language (EFL) writing course in a Hong Kong secondary school. The participants (99 Grade-10 students) were divided into treatment or control groups; the former and latter respectively used ChatGPT and conventional media as the tool for EFL writing instruction. The analysis of the participants’ EFL writing performance at the end of the experiment shows that the treatment group outperformed the control group. Further, there was a significant interaction effect between the assigned group of participants and their baseline level of EFL proficiency. Compared to the high-achieving participants, low-achieving ones tended to benefit more from ChatGPT-supported EFL writing instruction in the experiment. These findings together suggest that ChatGPT has the potential to be applied in formal EFL writing instruction in K-12 classroom settings.

The combination of playing and learning is a pivotal feature but also a key challenge in educational games. To investigate whether playing and learning are trade-off or complementary activities, this study explores students' distinction and transition between game behavior and learning behavior in a math spatial educational game for cube nets learning. A total of 402 fourth graders used the educational game over a 3-week period, with one lesson per week. Behavior log data, questionnaire data, and interview data were collected. The results show that the students could distinguish between game behavior and learning behavior mainly based on the related content of the behavior and their perceptions of the behavior characteristics. Two kinds of transition between game behavior and learning behavior were identified, namely the explicit large-scale cross transition and the implicit small-scale tight transition, which suggest a complementary relationship between playing and learning in educational games. Furthermore, compared with the students with lower learning outcomes, the students with higher learning outcomes were more likely to induce behavior patterns involving virtuous circles between playing and learning. The findings of this study have implications for effective educational game design going forward, in which playing and learning are appropriately combined.