Developing students’ technical competencies and self-regulated learning has become a priority in robotics and computing education, yet little is known about the mechanisms through which technical competence contributes to goal-setting behaviours in authentic learning environments. Drawing on Social Cognitive Theory and Goal-Setting Theory, we examined whether science-related attitudes mediate the relationship between technical competencies and robotics goal-setting skills and whether prior knowledge moderates this process. A total of 530 undergraduate students from Nigerian higher education institutions participated in a 15-week Arduino-supported Project-Based Learning intervention. Prior knowledge was assessed before the intervention, whereas technical competencies, science-related attitudes, and robotics goal-setting skills were measured concurrently at post-intervention. The hypothesised first stage moderated mediation model was analysed using Partial Least Squares Structural Equation Modelling with the SmartPLS PROCESS module and examined in a complementary robustness analysis using Hayes’ PROCESS macro in SPSS, with 10,000 bootstrap resamples. Results showed that technical competencies significantly and positively predicted robotics goal-setting skills and science-related attitudes. However, science-related attitudes did not significantly predict goal-setting skills or mediate the relationship between competence and goal setting. Prior knowledge did not significantly moderate the relationship between technical competencies and science-related attitudes, and the conditional indirect effect was unsupported. These results refine the joint application of Social Cognitive Theory and Goal-Setting Theory by suggesting that technical competence is associated with goal setting primarily through a direct competence-based pathway rather than the proposed attitudinal mechanism. This study underscores the importance of combining hands-on, feedback-rich robotics experiences with explicit goal-setting and self-regulatory strategies in higher education.
The rapid advancement of artificial intelligence has made AI-based instruction, including intelligent tutoring systems, learning analytics dashboards, adaptive learning platforms, and generative AI, increasingly prevalent in K-12 education. However, AI-generated information does not automatically lead to pedagogical action; it gains instructional meaning only when teachers interpret, judge, and translate it into situated classroom support. Following PRISMA guidelines, this systematic review analyzed 29 peer-reviewed English-language studies published between 2016 and 2025 across six databases to examine how teacher intervention is constituted in K-12 AI-based instruction and what conditions shape its reported effects. The synthesis was organized around three dimensions: process, strategy, and effect. Findings indicate that teacher intervention operates as a cyclical process of monitoring, judgment, intervention, and orchestration, through which AI-generated information is transformed into pedagogical action and subsequent classroom adjustment. Three broad intervention strategies were identified: pedagogical translation of AI outputs, design of learning support, and reconstruction of interaction structures. Reported benefits for student learning and teacher orchestration were generally promising but conditional, varying according to the interpretability of AI information, intervention timing, target level, teachers’ implementation feasibility, and students’ autonomy. By presenting an integrated conditional framework, this review argues that the teacher’s role in AI-based instruction is being reconfigured rather than diminished, offering implications for teacher professional development and the design of teacher-facing AI systems.
Programming is a fundamental tool for developing and applying computational thinking skills (CTS), which are closely linked to it and essential for individuals in the future. In this context, this study aims to model the variables associated with university students’ programming-oriented computational thinking skill (POCTS), with separate modeling for those experienced in scripting and programming languages. In this study, the correlational survey model was used and it was conducted with the participation of 426 university students. Personal information form, three different scales and semi-structured interview form were used as data collection tools. Data were analyzed by PLS-SEM and content analysis. As a result of the research, the relationships between CTS and academic self-efficacy, cognitive flexibility, academic achievement perception, extracurricular programming time and programming experience are significant. On the other hand, the direct effects of variables such as gender, department and education level are not statistically significant. It was observed that there were variations in the modeling in groups formed by the students who are experienced in programming languages and scripting languages. It was also found that personal variables did not affect all dimensions of POCTS. It was determined that both academic self-efficacy and cognitive flexibility significantly predict all dimensions of POCTS. However, cognitive flexibility demonstrated slightly stronger and more consistent effects across dimensions compared to academic self-efficacy.
Abstract Evaluations of educational technology (EdTech) increasingly demonstrate that observed outcomes are shaped not only by the design of an intervention, but also by the extent to which the conditions required for implementation are adequately accounted for. Despite advances in implementation science, the field still lacks a standardized and holistic framework for integrating these conditions into EdTech evaluations. This gap leaves providers, researchers, and funders without clear guidance on which implementation dimensions are critical to explain, support, and incentivize for effective use in real-world contexts. Our paper addresses this gap by drawing on a systematic review of the academic literature, textual analysis of keywords and consolidation of criteria with expert review, to harmonise implementation-related criteria into a conceptually grounded set of benchmarks that are operationalized as measurable and traceable indicators, which together constitute an EdTech Viability Index. Operationalized through a rubric and corresponding measurement guide, the Index is structured around four core dimensions: technical, economic, human, and system viability. These dimensions complement earlier work by a global research consortium on the 5Es framework of educational impact by extending the efficacy, effectiveness, ethics, equity, and environmental impact evaluations beyond demonstrated effects to the conditions under which those effects can be realized in practice. By integrating evidence of educational impact with evidence of implementation viability, the Index provides the first research-derived set of criteria for calculating and interpreting EdTech impact as both an outcome and a function of contextual readiness. The paper offers a measurement-oriented instrument through which the viability of interventions can be systematically assessed, documented, and followed over time.
Abstract Augmented Reality (AR) and Game-Based Learning (GBL) are emerging as transformative tools in education, particularly for developing computational thinking (CT) skills among middle school students. This study investigates the integration of AR-based GBL with eye-tracking technology and Epistemic Network Analysis (ENA) to evaluate learning outcomes and cognitive engagement. The study was conducted with 67 middle school students using a pre-test/post-test design. Participants engaged with an AR programming game specifically designed to foster CT skills. They used a marker-based AR educational game and game cards to solve programming challenges across 12 levels. An eye-tracking device recorded gaze fixation patterns, revealing students’ focus areas during gameplay. Quantitative analysis of the eye-tracking data, combined with ENA, highlighted differences between students with higher prior CT skills (HPCT) and lower prior CT skills (LPCT). HPCT students demonstrated strategic engagement, concentrating on advanced game levels and instructional content, whereas LPCT students primarily focused on foundational stages. Qualitative feedback analyzed with ENA from open-ended questionnaires provided additional insights. HPCT students offered constructive, well-justified feedback, indicating a deeper understanding of the educational potential of the AR environment. In contrast, LPCT students’ responses—though generally positive—lacked specificity, reflecting their difficulties with higher-order problem-solving. ENA revealed distinct patterns in how the two groups processed information and articulated feedback, underscoring the need for personalized instructional design. This research contributes a robust methodological framework by leveraging multimodal data to assess AR’s impact on CT development. Integrating quantitative and qualitative data through ENA in AR-GBL contexts can inform the design of more adaptable and inclusive learning environments. Future studies should examine the scalability of these findings across diverse populations and investigate the long-term effects of AR-GBL interventions on skill retention.
The integration of Artificial Intelligence (AI) into higher education promises to personalize learning, yet its rapid proliferation has outpaced the rigorous empirical evidence needed to guide its use, especially in complex humanities disciplines. The study of foreign literature, with its dual challenges of linguistic and cultural barriers, presents a high-stakes context where AI support is theoretically valuable but empirically unverified. This study, therefore, evaluated the effectiveness of a bespoke AI Digital Teacher designed to mitigate these challenges. In a semester-long randomized controlled trial, eighty-four undergraduate literature students were assigned to either a control group (traditional instruction) or an experimental group (traditional instruction plus AI teacher access). Outcome measures included academic performance, achievement emotions, and cognitive load. The results demonstrated a profound impact. The AI group significantly outperformed the control group on both objective tests (M = 81.9 vs. 75.2, p = .003) and, most notably, on analytical essays (M = 83.5 vs. 71.3, p < .001). Furthermore, a significant group-by-time interaction revealed diverging emotional trajectories: the AI group sustained positive emotions including enjoyment while decreasing negative emotions such as anxiety, whereas the control group exhibited opposite trends. Post-intervention, while intrinsic load was comparable, the AI group reported significantly lower extraneous load (M = 3.2 vs. 5.4, p < .05) and significantly higher germane load (M = 8.1 vs. 5.9, p < .05). These findings provide strong evidence that a well-designed AI Digital Teacher can not only substantially improve academic outcomes but also foster a more positive affective environment and optimize cognitive processing for deep learning in a complex humanities domain.
This study investigates the performance and fairness of predictive models designed to predict data science programming performance using fine-grained log data from a sample of students solving Data Science tasks in DaTu. Our analysis reveals that Logistic Regression outperformed other models in terms of accuracy. Through Variable Importance analysis, we found that the 'paste answer' variable emerged as a critical predictor for Logistic Regression and KNeighbors classifier models, whereas multiple variables contributed equally to the GaussianNB model's predictive power. We evaluated fairness, focusing on the variable 'chatbot access' (with or without access to chatbot), finding evidence of bias. The models demonstrated lower score distributions for the unprivileged group (no-AI access). To mitigate this bias, we employed various techniques: Resampling, Reweighting, and ROC pivot. Of these, only Reweighting significantly improved fairness metrics. As an alternative approach, we considered the ceteris paribus cutoff method to minimize parity loss. In conclusion, this study emphasizes the significance of evaluating fairness in educational predictive modeling. This work offers a nuanced and data-driven approach to assessing students' data science programming skills.
The regulation of learning, including self-regulated learning (SRL) and socially shared regulation of learning (SSRL), is a key skill that should already be practiced in primary education. The shift toward student-centered, collaborative, and open-ended processes, such as phenomenon-based learning, increases the need for pupils to take charge of their own learning. However, due to the varying abilities of primary school pupils to regulate their learning, such processes can be challenging and require substantial scaffolding. To address this issue, this study explores how a learning design that indirectly scaffolds the regulation of collaboration can support regulation during the challenging phases of a collaborative blended learning process in primary school. Additionally, the potential of learning analytics (LA) to further support pupils’ processes is discussed. A specific study module on sustainable development was designed and implemented in a learning management system (LMS). Classroom observations and LMS log data were collected during the implementation with fifth- and sixth-grade pupils. The observation data show that the most challenging phases of the group work occurred at the beginning of the group process and again during the finalization of the group project. The log data further indicate that pupils used the LMS as a regulation support at the beginning of the process. Overall, the results highlight the need for more functional tools that help groups to metacognitively monitor and summarize their learning. In addition, LA could provide meaningful visualizations for teachers to further develop learning designs during learning processes.
With the development and integration of information and communication technologies, the learning needs and skills have revised and refined, and 21st-century digital skills have recently become more important than before. Computational thinking skill, the heart of the 21st-century digital skills, has recently been developed and validated in the field of computer-assisted language learning (CTCALL), but has received little attention in CALL. Consequently, the researcher cultivated this important skill in Intelligent CALL (ICALL) for 104 Iranian EFL learners and explored it on the language learners' motivation and their learning approaches to ICALL. The bisymmetric research design validated the factorial validity of this skill and showed that the abstraction and evaluation skills were among the necessary conditions to shape language learners' deep approach to ICALL, and further level of them can shape their deep approach to it as well. However, generalization was not among the necessary conditions, but the higher language learners can identify patterns in solving specific language tasks with AI, and applying those patterns to other language tasks involves new ways of thinking and cross-referencing ideas with AI, culminating in maximizing their ICALL for learning language. Based on these findings, the study provides a new conceptual model to the field of CALL and recommends that teachers adopt a problem-solving approach within ICALL and cultivate new skills, such as CTCALL, in their language learning environment to foster motivation and deep, organized approaches to ICALL.
Educational recommendation systems have traditionally relied on single-dataset approaches, limiting their ability to capture the complex, multi-faceted nature of student learning. This paper introduces a novel multi-modal graph neural network framework that integrates heterogeneous educational data sources to deliver superior personalized learning recommendations. Our approach combines behavioral learning analytics from EdNet with institutional context from OULAD, creating a large-scale cross-dataset educational framework. The proposed architecture employs Graph Convolutional Networks for structural modeling, Graph Attention Networks for dynamic weighting, and hierarchical temporal components to capture learning dynamics. Novel cross-modal attention mechanisms enable knowledge transfer between behavioral patterns and contextual factors, while cognitive load-aware optimization ensures educationally appropriate recommendations. Comprehensive experimental evaluation demonstrates substantial improvements in recommendation accuracy and educational effectiveness. Individual-level assessment reveals high accuracy in predicting students’ actual learning choices, with superior success rates for recommended learning activities. Cross-dataset transfer learning achieves excellent performance, showing significant improvements over traditional domain adaptation approaches. Beyond performance metrics, our framework delivers tangible educational benefits including substantial reduction in learning time while maintaining high engagement levels through adaptive optimization. The system demonstrates its capability in learning gap identification and targeted remediation, with strong correlations to educational psychology indicators validating pedagogical authenticity.
Multimedia has been recognised as a powerful domain for contextualising programming concepts. However, the inherent complexity of multimedia applications, particularly their reliance on advanced data structures, often poses significant challenges for novice programmers. To address this issue, we implemented Medialib, a user-friendly Python multimedia library specifically designed for beginners at or above the high school level. Medialib was developed through an iterative process informed by empirical studies involving non-technical university students and their instructors, with the goal of making multimedia programming more accessible to learners without prior technical backgrounds. This paper introduces, for the first time, a simplified multimedia Python library and accompanying pedagogical materials tailored to the cognitive and instructional needs of novice programmers. Medialib enables a pedagogical shift in introductory Python courses from traditional mathematics-oriented exercises to multimedia-focused tasks. To evaluate its effectiveness and transferability, two empirical studies were conducted: a 14-week study in Japan (21 instructional hours, 36 students) and a 2-week study in the UK (12 instructional hours, 84 students). Analyses are done on the study data which includes teacher observational notes, questionnaires, and interviews. Specifically, a comparison of the weekly performance of learners in traditional maths-related exercises and Medialib-related exercises in the first study is discussed. Findings from both studies indicate that learners responded positively to the Medialib materials. Notably, in the Japanese study, students who initially struggled with maths-related programming tasks were able to successfully acquire foundational programming skills through Medialib activities. From week 7 onward, students consistently demonstrated strong performance in both types of exercises, suggesting that Medialib serves not only as an effective entry point for programming education but also as a transferable learning scaffold across contexts.
This study investigates the effects of AI-supported online classroom management (AIS-OCM) on students’ attitudes toward English as a Foreign Language (EFL), their self-directed online learning skills, and academic achievement. The rapid development of digital technologies, particularly in the aftermath of the COVID-19 pandemic, has highlighted the need to sustain student engagement and motivation in online learning environments. Within this framework, AI-based tools and intelligent learning analytics systems offer promising potential for delivering personalized feedback and adaptive learning support. A quasi-experimental design was adopted, involving 7th-grade students from an EFL context. While the experimental group received AI-supported online instruction over eight weeks, the control group participated in standard online lessons without AI integration. Data were collected using pretest and posttest measures and analyzed with Mixed ANOVA to examine time, group, and time × group interaction effects focusing on attitudes toward EFL lessons, self-directed online learning, and academic performance. The findings indicate positive learning gains over time for both groups and generally more favorable outcomes for the AIS-OCM group, offering meaningful insights into the multidimensional impact of AIS-OCM. These results are expected to contribute to the growing body of research on AI-supported education and provide practical implications for educators and policymakers operating in EFL settings.
Abstract With several fragmented literature reviews and meta-analyses on Large-Language Models (LLMs) in education, this study provides a synthesis of these reviews focusing on the role of LLMs to facilitate different pedagogy paradigms to reveal research trends, existing gaps, and future directions. The synthesis adheres to the PRISMA guidelines and AMSTAR checklist to analyze 50 reviews to find out the trends of research in terms of publication year, geographic regions, types of reviews, types of research questions, pedagogy paradigms addressed, and challenges encountered. Findings revealed that constructivism, cognitivism, and connectivism emerged as the most frequently addressed paradigms, indicating the role of LLMs as (i) tools for learning discovery or scaffolding, where learners actively construct knowledge, (ii) tools to improve cognitive tasks (e.g., recall, comprehension, problem-solving), or (iii) tools to facilitate connections between diverse knowledge sources or enable networked learning environments. Additionally, several challenges emerge, which are related to cognitive load and processing, accuracy and comprehension, cultural sensitivity and diversity, learner autonomy and self-expression, among others. This study offers valuable insights into the evolving roles of LLMs in facilitating paradigms that contribute to a deeper understanding of the pedagogical implications, challenges, and opportunities presented by LLM adoption in education.
Abstract English language proficiency is essential for academic and career prospects of individuals who speak English as a second language (ESL). Yet, gaining English proficiency is a difficult task for ESL learners, particularly adults who may not have acquired the language in their formative stages. Gamification helps to bridge the gap by making language learning enjoyable. But there is no clear understanding of which strategy works best for ESL adult learners between personalized (individual-based) and collaborative (group-based) gamification. To address this gap, this study presents a systematic synthesis of the literature on gamified vocabulary learning among adult ESL learners. This review included 19 articles published between January 2015 and December 2024. The findings show that gamification, irrespective of whether it is personalized or group-oriented, has beneficial impacts on language learning, engagement, and motivation. Individual-based gamification promotes learner autonomy while collaborative learning encourages social interaction. Both strategies work for the benefit of learners’ academic achievement, motivation, and engagement. Learning contexts seemed to influence the success of these strategies. The findings indicate that teachers have to weigh the contexts of their ESL classrooms in the design of gamification strategies.
Learning engagement is inherently dynamic, historically dependent, and oriented to long-term utility. Building on sequential decision-making theory, this study simulates the dynamic, long-term evolution of online learning engagement among Chinese EFL learners using reinforcement learning models for the first time. The findings unveil that online learning engagement can be reconceptualized as a sequential decision-making process. In the process, three optimal policy regimes, stable reinforcement strategies, conservative strategies, and proactive strategies, are found. They can explain how learning engagement is sustained, stabilized, or redirected across various learners’ profiles. Furthermore, learner attributes such as online self-efficacy, metacognitive engagement, and social interactions, jointly reconfigure the sequential decision-making process, altering state representations, rewards, and action costs, thereby generating divergent optimal decision-making strategies. Therein, social interaction is the most powerful predictor. Despite its intrinsic limitations, the present study underscores the essential nature of language learning. More critically, it proposes an innovative interdisciplinary framework for SLA research, which synthesizes insights from decision science, cognitive science, machine learning, and SLA itself.
Abstract As Artificial Intelligence (AI) has been widely used in education, understanding older adults’ perceptions and acceptance of AI-supported learning is crucial for enhancing technology access and promoting lifelong learning. This study analyzed 75 older adults from diverse gender, education, and technical experience, collecting both hand-drawn and AI-generated images. These images were coded into six categories with 20 elements. Drawing-based Epistemic Network Analysis (ENA) was employed to analyze the results, while interviews provided supplementary insights into their perceptions. The findings revealed that (1) hand-drawn works reflected older adults’ life-based understanding, while AI-generated drawings contained more structured elements; (2) gender, education, and technical experience influenced the content and style of the drawings: males focused on self-improvement, females on emotions like parent–child relationships, those with higher education created more organized scenes, while those with lower education emphasized emotional expression, and more tech-savvy individuals were exploratory, while others required assistance; and (3) most older adults found AI tools practical and expressed willingness to use them. These results highlight key factors influencing technology adoption among older adults and underscore the need for educators to consider contextual variations in AI-supported learning.
The integration of artificial intelligence (AI) in education is prompting a reevaluation of personalized learning terminology and its impact on teaching practices and learner engagement. Personalized learning (PL) involves various instructional strategies tailored to individual student needs and interests, utilizing data and technology to boost engagement and success. The evolving landscape requires a clear understanding of how AI can support personalized learning, distinguishing it from traditional methods. The variability in PL terminology reflects diverse interpretations of AI technologies in education, necessitating a common framework to clarify definitions and practices. This document presents an overview of the latest research literature on personalized learning, highlighting how technology is transforming the framework and effectiveness of individualized learning experiences. By analyzing reputable articles from 6 databases, the review seeks to provide insights into how AI can redefine personalized learning, enabling more precise definitions. The findings emphasize the use of PL terms in technological contexts and call for a unified term to enhance clarity and effectiveness in educational technology practices. Ultimately, the review aims to inform educators and policymakers about precise terms defining personalized learning in the AI context.
Massive Open Online Courses (MOOCs) have widened access to education globally; however, they continue to face course overload, limited guidance, insufficient context-aware personalization, and high dropout rates. These challenges are even more pronounced for learners with disabilities, due to limited accessibility support and the absence of recommendation systems that adapt to individual accessibility needs. This study introduces RECMOOC4ALL, an AI-enhanced, hybrid multi-signal recommender system that embeds accessibility as a first-class computational feature within its ranking logic. Guided by Universal Design for Learning principles, the system integrates content-based filtering, neural collaborative filtering, sentiment analysis, and dropout-risk modeling, while encoding course metadata such as caption availability, screen-reader compatibility, and keyboard navigability. Evaluation using a multi-platform dataset (11,600 courses, 42,000 interaction sessions, and 35,000 textual reviews) shows that the accessibility-aware hybrid achieves lower error (RMSE 0.80, MAE 0.63) and higher ranking quality (Precision@5 0.33, Recall@5 0.36, NDCG@5 0.34) than CBF and CF/NCF baselines. A user study (n = 50) reported satisfaction of 4.28 ± 0.47 and accessibility satisfaction of 4.36 ± 0.39, and log-based analyses indicated increased click-through, longer dwell time, and higher course completion for recommendations produced by RECMOOC4ALL compared with baseline configurations. These gains were achieved while prioritizing accessibility-compliant courses via calibrated ranking weights. RECMOOC4ALL demonstrates that accessibility can be operationalized at the computational core of recommendation, offering a scalable and empirically validated blueprint for equitable AI in MOOCs.
This research provides a systematic review and comparison of 14 existing Open Educational Practices (OEP) models and frameworks. Although OEP has its origins in Open Educational Resources (OER), the theoretical conception of OEP remains disparate and heterogeneous. In order to address this issue, the study operationalizes the holistic view of the models by providing a multi-dimensional comparison of these models across their focus, key players and stakeholders, core processes and practices, values and technology orientations. The empirical results point at the holistic evolutionary trajectory for OEP: a Resource Dimension focused on access and creation. A Teaching Dimension focuses on open pedagogy and learner agency. A Technology Dimension focuses on technology, shifting roles from mere tools to integral pieces of educational infrastructure. In line with this development, this study proposes a novel AI-empowered Open TPACK Framework as the next stage of evolution of OEP, where AI is conceptualised not as a tool but as the infrastructure of Content, Pedagogy, and Technology Knowledge (TPACK). This study asserts that the transition from “Resource Access” to “AI Empowerment” indicates a meaningful change in the concept of Open Education. This research provides direction and guidance to educators and policy-makers in effective and meaningful OEP in an algorithmic age, highlighting the need for human agency within clever ecosystems to create sustainable learning environments.
