Large Language Models (LLMs), such as ChatGPT, Claude, and Gemini, have been widely adopted, significantly influencing various domains of social life, particularly education. This work highlights the importance of evaluating students' cognitive processes of learning-such as information gathering, decision-making, and assumption formation-when interacting with LLMs. We propose a BERT-based method using Bloom's Taxonomy to analyze students' task prompts and gain insights into their reasoning and problem-solving. Analysis of 48 students over a 16-week Python Programming course at a national university in northern Taiwan reveals a preference for higher-order cognitive skills. Application prompts comprise 32% of total prompts, compared to 12% for Knowledge. DeBERTa achieved the highest accuracy (75%) and F1 score (0.729)., though error rates were higher for complex tasks like Synthesis (0.44) than for Knowledge (0.13). These findings demonstrate the potential of LLMs to enhance critical thinking and support targeted educational interventions. Our findings indicate that LLMs facilitate content mastery and enhance awareness of how students approach educational tasks. By incorporating data from LLM interactions into assessment practices., educators can better understand learners” competencies., enabling targeted interventions to develop critical thinking skills. This approach provides a foundation for more robust pedagogical frameworks., ensuring students' growth through dynamic and reflective learning experiences.
Pedagogical Conversational Agents (PCAs) are increasingly used to provide personalized support in digital learning. This study examines how different levels of learner control that a virtual teaching assistant (TA) offered impact students' self-efficacy, learning effectiveness, and satisfaction, given the moderating effects of student characteristics. A between-subjects study with 215 students across three learner control conditions revealed that the moderate control (i.e., Assisted Inquiry) led to higher perceived learning effectiveness than the high control. Agreeableness, Extraversion, and Openness were found to significantly influence students' learning. These findings underscore the need for adaptive PCA systems that tailor instructional support with scaffolding and balanced learning interventions while accounting for learners' personality traits to enhance student engagement and perceived learning effectiveness.
In recent years, namely due to its association with positive academic outcomes, the concept of ‘student engagement’ has received a growing interest from researchers, practitioners, and policymakers. As student engagement received strong attention due to its association with positive academic outcomes, research on the costs and benefits of technology use has also become increasingly popular. This paper presents an exploratory literature review aiming to map how student engagement is being assessed in TEL environments, highlighting the interplay between self-reported experiences, learning analytics, and ethical considerations. The potential benefits and risks associated with engagement measurement in TEL are also addressed. The review identifies key gaps in current research and suggests directions for future exploration, particularly at the intersection of student engagement, TEL, digital well-being, and ethics.
In this study, we examined the effects of a 1.5-day microelectronics training program on students' self-efficacy. Using a quasi-experimental design (N = 27), we analyzed the data through a paired t-test complemented by qualitative insights. The findings revealed that: (1) a short-term microelectronics training program, designed based on the Attention, Relevance, Confidence, and Satisfaction (ARCS) motivational model, can positively influence learners' self-efficacy toward STEM education; and (2) incorporating diverse instructional strategies-such as guest talks from industry and academic experts, along with hands-on, real-world experiences-is essential for making abstract textbook concepts more concrete and inspiring students to pursue STEM-related careers.
This study investigates iVRLab, an immersive virtual reality (VR) microfabrication training system, and its influence on college students' self-efficacy and interest in microelectronics during a one-hour workshop held within a 1.5-day training camp. iVRLab simulates photolithography cleanroom operations, providing hands-on virtual experiences that merge theory with practice. A pre-post analysis revealed a significant rise in self-efficacy (M = 4.11 to 4.37, p = 0.0066), demonstrating the system's effectiveness in building confidence. Although interest only showed slight gains, high baseline scores (3.6-3.7 on a 4-point scale) indicate a ceiling effect. Correlations among engagement, embodiment (r = 0.91, p < 0.001), and immersion (r = 0.60, p < 0.01) underscore the value of active, embodied VR learning. Qualitative feedback further highlighted iVRLab's immersive realism and its capacity to spark curiosity about microfabrication careers. These findings suggest that VR training can effectively boost self-efficacy in microelectronics, reinforcing students' enthusiasm in an informal learning setting and indicating the potential broader adoption across STEM fields.
Profiling readers provides valuable insights into the interaction between affective and cognitive engagement in multiple-text reading. However, metacognition-particularly goal-oriented and problem-awareness metacognition-remains underexplored. This study integrates unsupervised machine learning (k-means clustering) and structural equation modeling (SEM) to examine how these metacognitive factors interact with affective engagement (topic interest and self-efficacy) and their relationship with reading processes and writing performance. A total of 152 participants read four texts on genetically modified technology and wrote a synthesis report. Their reading behaviors were recorded using log data and eye-tracking technology to capture the integration process, while writing performance was assessed through multiple-text integration and source citation. Results revealed three distinct reader profiles: high-metacognition, high-interest, and disengaged groups. SEM analysis showed that, compared to the high-interest group, the high-metacognition group exhibited greater integration behaviors, which in turn enhanced multiple-text integration and source citation. Moreover, the integration process successfully mediated the relationship between reader profiles (High-metacognition vs. High-interest) and writing performance. These findings highlight the critical role of goal-oriented metacognition in multiple-text reading and emphasize the need to further explore metacognitive factors in comprehension research.
The exponential growth of Massive Open Online Courses (MOOCs) has expanded educational access but has also overwhelmed learners with an excessive number of choices, making it difficult to identify courses that align with their skills, needs, and goals. To address this challenge, we introduce RecMOOC4All, an AI-powered, accessible MOOC aggregator that centralizes course discovery and intelligently recommends personalized learning pathways. At its core, RecMOOC4All integrates a hybrid recommender system, combining content-based and collaborative filtering to suggest courses based on learner profiles, preferences, accessibility requirements, and past interactions. An automated accessibility checker and remediation module ensures WCAG compliance, incorporating real-time captions, alternative text, and enhanced navigation to foster inclusivity. Furthermore, a generative AI-powered conversational chatbot provides real-time, context-aware assistance, while engagement detection modules analyze learner behavior to trigger timely interventions, improving retention and learner success. By seamlessly integrating personalization, accessibility, and engagement tracking, RecMOOC4All creates a cohesive, learner-centric MOOC environment, towards inclusive, adaptive, and effective education is for all learners.
This study examined graduate students' attitudes toward ChatGPT in thesis writing and its role in producing the first three chapters. Students positively noted its efficiency and structural support while acknowledging limitations like inaccuracies. Expert evaluations showed a positive correlation between writing quality and originality. The findings highlight the need for responsible use, critical engagement, and training to maximize ChatGPT academic potential.
This paper explores the development and preliminary evaluation of a simulation system designed to train nursing students to handle cardiac arrest situations. The simulation, developed using Unity for Meta Quest devices, creates a high-fidelity immersive environment where users interact with multiple patients, monitor vital signs, and respond to emergencies. The system incorporates continuous alarms, auditory cues, and haptic feedback to replicate real-world stressors. A pilot study conducted evaluated the system's feasibility as an educational tool through pre- and post-assessment surveys. The evaluation focused on user experience, engagement, and cognitive load in simulated high-pressure medical scenarios. The results demonstrate that immersive virtual reality technology can provide a secure and efficient environment for enhancing clinical decision making skills while maintaining psychological fidelity.
In service-oriented disciplines like hospitality and nursing, multimodal knowledge such as gestures, movements, and facial expressions is crucial for professional performance. Extended reality learning environments provide effective environments for situated learning, yet controller-based interactions often fail to cultivate authentic muscle memory. Building on embodied cognition and multimedia learning principles, this study develops a digital learning approach that integrates physical object manipulation, sensory-motor feedback, and AI-driven multimodal assessment to enhance skill acquisition. Using a 3D stereo camera and AI recognition, the system constructs digital twins of learners and real objects, synchronizing physical and virtual performances. By shifting from an operator to an audience perspective, learners can observe their own performance in a virtual scene and receive real-time AI-driven multimodal assessments to enhance reflection and knowledge internalization. Additionally, incorporating physical object manipulation instead of controller-based interaction provides authentic sensory feedback, reducing cognitive load and strengthening the link between cognition and motor execution. Experiments conducted in a practical hospitality course at a university of science and technology demonstrate that, compared to controller-based manipulation of virtual items, this approach improved learning outcomes and reflection with lower cognitive load, highlighting the advantages of integrating multimodal feedback and physical objects in digital situated learning.
This paper presents the design and implementation of a student-facing Learning Analytics Dashboard (LAD) to support collaborative Virtual Reality (VR) content creation in educational settings. The LAD includes a shared task checklist, class progress statistics, and visualizations of student contributions to promote self-regulated learning (SRL) strategies and group awareness. The effectiveness of the LAD was evaluated through a quasi-experiment in a high school VR creation program with a total of 111 students. Results revealed high adoption rates of the LAD and significant differences in editing behaviours between the control and experimental groups. This study demonstrates the LAD's potentials in fostering iterative improvements and enhancing collaboration in collaborative maker activities.
This study aimed to enhance junior high school students' artificial intelligence (AI) literacy through the practical application of an AI-based instructional support tool developed for school education. The tool was designed to facilitate idea generation and thought organization by allowing students to input questions and receive AI-generated responses within predefined learning tasks and constraints, addressing concerns about output quality and task specificity in generative AI. A fifty-minute lesson was conducted with 175 students at a junior high school in A Prefecture, Japan. The lesson introduced the basic concepts of generative AI, reviewed usage guidelines, and provided hands-on experience with the tool in the context of exploring ways to promote their city. Post-lesson survey results indicated improvements in certain aspects of AI literacy. However, the findings also highlighted the need for future instruction to incorporate methods for assessing the reliability of AI-generated outputs.
Prediction of student performance is one of the main research lines in learning analytics, whose aim is to detect students at risk to support them. Nevertheless, current literature often relies on a single platform to analyze data. However, some courses use several platforms and it is relevant to know which type of platform/resources is more relevant for prediction. In this line, the objective of this work is to analyze the effect of multiple sources when conducting performance predictive models. Particularly, data are collected from a spreadsheets course where there is information about activity in the Learning Management System (LMS, Moodle in this case), a supporting Small Private Online Course (SPOC), class attendance registered in the Blackboard Collaborate online tool, and academic results. Results show that it is possible to obtain accurate predictions in both final test and final grade, and Moodle interactions and academic results stand out for final grade predictions, while class attendance activity is a strong predictor of the final test. Conversely, the SPOC formative interactions show a worse predictive power in comparison to other variables, which may reflect that students do not use the supporting videos if they grasp the contents using the materials provided in Moodle.
Business English has become essential skills for information-majored students to succeed in the global workplaces. In such environments, effective communication is not only a competitive advantage but often a fundamental requirement for collaboration and problem-solving across diverse cultural and professional contexts. However, traditional learning methods may fall short in addressing the unique linguistic and practical needs of these students. In this regard, artificial intelligence (AI) emerges as a game-changer in modern education, offering innovative solutions to enhance learning experiences. AI's ability to deliver personalized and interactive instruction allows individuals to focus on areas where they need the most improvement and also maintain the engagement.
Activities that engage learners to articulate their answers often make them reflect. However, evaluating such activities and providing feedback is time-consuming for teachers. For text analysis, various data-driven indicators, such as cohesion and coherence, evaluate linguistic measures and the semantic understanding of artefacts created. However, for drawing-based activities, defining such indicators is still underexplored. In this research, we conducted a draw-and-write activity that engaged students to express their understanding of a concept through writing and drawing. The question was, “What is data science?”. The human raters analyzed the artefacts generated (n=40), and then a learning analytics approach was taken to define data-driven indicators. The study proposes a data processing pipeline involving a large language model (LLM) and defines indicators to understand the coherence of written text and drawn diagrams. Further, a clustering analysis of the collected artefacts highlighted differences in the participants' expressions of data science (task context). The discussion compares automated and human classification and its implications for assessment and feedback. Future work aims to integrate the pipeline in an online learning environment that affords drawing and text input from the learners.
Adaptive Learning Mechanisms (ALM) can be used to personalize the learning process in Learning Management Systems (LMS) to increase learning gains. The issue is that existing approaches to integrating ALMs into LMSs often do not focus on supporting the user groups that are interested in specifying and providing ALMs (i.e., instructors and researchers) according to their needs. This is because these approaches do not focus on specifying the provided ALMs, require a high level of technical expertise to specify them, or are not intended to integrate the provided ALMs into different LMSs to work with existing content. With the goal of addressing these requirements, we propose a concept of an LMS-independent modeling interface. This interface allows users with a low level of technical expertise to model ALMs, referred to as assistance workflows, in a flowchart-like graph. These ALMs can work with different LMSs based on a service-oriented architecture. We evaluated the proposed concept by means of a technical evaluation and a user evaluation according to the user-centered design approach. The results indicate that the proposed architectural approach is suitable to be applied to different LMSs and that the modeling interface meets the requirements of the target user groups.
This study explores the integration of Music Information Retrieval (MIR) techniques into learning analytics to analyze students' background music choices during virtual reality (VR) content creation. Through a mixed-methods approach involving interviews with 16 students and the analysis of background music in 98 VR stories, we examined students' music selection strategies and the emotional and stylistic characteristics of their chosen audio. Findings reveal a preference for calm, low-arousal music aligning with cultural heritage themes. The study demonstrates the potential of MIR in educational contexts, contributing to the emerging field of multimodal learning analytics.
Most classroom educational technologies adopt deficit approaches, which can be detrimental, particularly for historically marginalised students. To counter these approaches, we investigate the feasibility of leveraging ‘Funds of Knowledge and Identity’ (FoK/I) within Knowledge Tracing (KT) tools.
With the rapid advancement of generative AI (GenAI), collaborative learning between humans and machines has emerged as a growing field in education. However, further exploration of its efficacy is needed. This study examines the impact of transparent GenAI on learning outcomes, focusing on a human-machine trust model. It investigates whether varying transparency levels, defined by an agent's ability to explain its decisions, influence learner trust and, in turn, learning outcomes. GenAI agents were developed using a Performer-based deep learning architecture to identify errors in students' object-oriented programming (OOP) code. Nineteen participants were involved in a randomized block experiment evaluating trust, cognitive load, and learning performance with two agents of different transparency levels. A paired Wilcoxon signed-rank test revealed significant differences in trust and exam scores, indicating that transparency enhances trust and learning outcomes, though no substantial difference in cognitive load was found. Additionally, SmartPLS Multi-Group Analysis (MGA) revealed a positive correlation between trust and learning outcomes, especially for the highly transparent agent, suggesting the mediation effect of trust on the relationship between transparency and learning outcomes.
Extensive research has demonstrated that the integration of technology into language education significantly enhances the writing skills of English as a Foreign Language (EFL) learners. However, limited attention has been given to the application of AI-driven video-to-text recognition tools for supporting structured writing tasks in authentic learning environments. To address these gaps, this study developed the Video-to-Text Recognition (VTR system), which combines AI and cloud-based technologies to help learners build vocabulary and write sentences through real-world tasks. A cohort of 22 first-year university students participated in a two-week experimental study designed to evaluate the impact of the VTR system on writing proficiency. The results showed improvement in writing skills, with students giving positive feedback about the system's ease of use, engagement, and practical benefits. This study shows that the VTR system is a valuable and scalable tool for improving EFL writing and highlights the importance of combining advanced technology with real-world tasks to make language learning effective.
