Pub Date : 2025-12-19DOI: 10.1109/TLT.2025.3646366
Bérénice Lemoine;Pierre Laforcade;Sébastien George
The long- and short-term memorization of declarative knowledge (DK) requires repetition, which can quickly lead to boredom for learners. Serious games are often developed to support this training by offering a better motivational vector. However, a key design challenge is providing learners with experiences that are consistently varied and adapted. Generating adapted activities automatically is a technique that is little discussed in the field of technology-enhanced learning. This work addresses the challenge of designing generators for adapted and varied activities within serious games dedicated to DK training. The core contribution is the proposal of a comprehensive software framework aimed at facilitating the design and implementation of these generators. This framework is built upon the theories and practices of model-driven engineering (MDE), utilizing models and metamodels to ensure formal specification, abstraction, and reusability. The framework is designed to be extensible to different didactic domains. It enables the production of training activities that are simultaneously varied and adapted, considering adaptation from both the educational perspective (learner's level and progress within a training path defined by teachers) and the gaming perspective (player's preferences about gameplay). The activities produced by generators developed using this framework take the form of Roguelite-oriented dungeon levels, a genre chosen because its inherent procedural generation provides the necessary variety, repetition, and progression characteristics for DK acquisition. This article presents the framework, including its conceptual models and its MDE-based software infrastructure. We demonstrate its ability to express different didactic domains through its application to distinct educational contexts. Finally, this article details the evaluation from an engineering point of view, including the use of automated system tests to verify that the generated activities satisfy the properties of adaptation and variety.
{"title":"Design and Implementation Framework of Game Activity Generators for Declarative Knowledge Training","authors":"Bérénice Lemoine;Pierre Laforcade;Sébastien George","doi":"10.1109/TLT.2025.3646366","DOIUrl":"https://doi.org/10.1109/TLT.2025.3646366","url":null,"abstract":"The long- and short-term memorization of declarative knowledge (DK) requires repetition, which can quickly lead to boredom for learners. Serious games are often developed to support this training by offering a better motivational vector. However, a key design challenge is providing learners with experiences that are consistently varied and adapted. Generating adapted activities automatically is a technique that is little discussed in the field of technology-enhanced learning. This work addresses the challenge of designing generators for adapted and varied activities within serious games dedicated to DK training. The core contribution is the proposal of a comprehensive software framework aimed at facilitating the design and implementation of these generators. This framework is built upon the theories and practices of model-driven engineering (MDE), utilizing models and metamodels to ensure formal specification, abstraction, and reusability. The framework is designed to be extensible to different didactic domains. It enables the production of training activities that are simultaneously varied and adapted, considering adaptation from both the educational perspective (learner's level and progress within a training path defined by teachers) and the gaming perspective (player's preferences about gameplay). The activities produced by generators developed using this framework take the form of Roguelite-oriented dungeon levels, a genre chosen because its inherent procedural generation provides the necessary variety, repetition, and progression characteristics for DK acquisition. This article presents the framework, including its conceptual models and its MDE-based software infrastructure. We demonstrate its ability to express different didactic domains through its application to distinct educational contexts. Finally, this article details the evaluation from an engineering point of view, including the use of automated system tests to verify that the generated activities satisfy the properties of adaptation and variety.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"19 ","pages":"1-20"},"PeriodicalIF":4.9,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145963441","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-02DOI: 10.1109/TLT.2025.3639317
Marco Zappatore;Minjuan Wang;Chris Dede;Xuefan Li
The modern educational landscape is frequently crossed and sometimes even deeply reshaped by novel digital technologies, which promise to disclose new paths for teaching, learning, and research. Among a significant number of engaging approaches, the concept of intelligence augmentation (IA) is emerging as one of the most transformative since it involves a complementary partnership between human and artificial intelligence (AI) [1]. The aim is as ambitious as intriguing, since human intelligence’s strengths (e.g., judgment, ethics, and practical knowledge) can be relevantly boosted when augmented with the tasks that AI typically excels at (e.g., computation, prediction, and data-driven analysis) [1].
{"title":"Guest Editorial: Special Issue Intelligence Augmentation and the Future of Education: Transforming Learning Landscapes Across Modalities and Lifecycles","authors":"Marco Zappatore;Minjuan Wang;Chris Dede;Xuefan Li","doi":"10.1109/TLT.2025.3639317","DOIUrl":"https://doi.org/10.1109/TLT.2025.3639317","url":null,"abstract":"The modern educational landscape is frequently crossed and sometimes even deeply reshaped by novel digital technologies, which promise to disclose new paths for teaching, learning, and research. Among a significant number of engaging approaches, the concept of intelligence augmentation (IA) is emerging as one of the most transformative since it involves a complementary partnership between human and artificial intelligence (AI) [1]. The aim is as ambitious as intriguing, since human intelligence’s strengths (e.g., judgment, ethics, and practical knowledge) can be relevantly boosted when augmented with the tasks that AI typically excels at (e.g., computation, prediction, and data-driven analysis) [1].","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1112-1115"},"PeriodicalIF":4.9,"publicationDate":"2025-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145830781","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-27DOI: 10.1109/TLT.2025.3637864
Chun Man Victor Wong;Yen Na Yum;Rosanna Yuen-Yan Chan
Applied behavior analysis (ABA) therapy is a widely used intervention for students with special education needs (SENs), particularly those with autism spectrum disorder and co-occurring intellectual disabilities. However, despite its proven effectiveness, the integration of artificial intelligence (AI) and learning analytics (LA) in ABA therapy remains largely underexplored. This study examines the impact of an AI-driven LA system on prediction performance, intervention effectiveness for SEN students, and support for therapists and teachers. The system collects and analyzes physiological, environmental, and behavioral data in real time to generate personalized intervention recommendations. A total of 33 students and 26 therapists/teachers from special schools and therapy centers in Hong Kong participated in an eight-week ABA intervention, followed by a postevaluation session. The study assessed predictive accuracy, student learning outcomes, and educator perceptions using empirical data and qualitative feedback. Results indicate that the system achieved a predictive accuracy of 88.83% and a precision of 86.64% in forecasting learning outcomes, with statistically significant student performance improvement (medium effect size). Educators reported that the system’s AI-driven recommendations enhanced their ability to develop individualized student profiles and intervention strategies. While the system did not replace traditional ABA methodologies, it improved decision making by providing actionable insights through multimodal data integration. As of today, our system has been used by over 1000 students with SENs in Hong Kong, Singapore, and Canada, demonstrating the real-world impact of AI-driven LA.
{"title":"AI-Driven Learning Analytics for Applied Behavior Analysis Therapy","authors":"Chun Man Victor Wong;Yen Na Yum;Rosanna Yuen-Yan Chan","doi":"10.1109/TLT.2025.3637864","DOIUrl":"https://doi.org/10.1109/TLT.2025.3637864","url":null,"abstract":"Applied behavior analysis (ABA) therapy is a widely used intervention for students with special education needs (SENs), particularly those with autism spectrum disorder and co-occurring intellectual disabilities. However, despite its proven effectiveness, the integration of artificial intelligence (AI) and learning analytics (LA) in ABA therapy remains largely underexplored. This study examines the impact of an AI-driven LA system on prediction performance, intervention effectiveness for SEN students, and support for therapists and teachers. The system collects and analyzes physiological, environmental, and behavioral data in real time to generate personalized intervention recommendations. A total of 33 students and 26 therapists/teachers from special schools and therapy centers in Hong Kong participated in an eight-week ABA intervention, followed by a postevaluation session. The study assessed predictive accuracy, student learning outcomes, and educator perceptions using empirical data and qualitative feedback. Results indicate that the system achieved a predictive accuracy of 88.83% and a precision of 86.64% in forecasting learning outcomes, with statistically significant student performance improvement (medium effect size). Educators reported that the system’s AI-driven recommendations enhanced their ability to develop individualized student profiles and intervention strategies. While the system did not replace traditional ABA methodologies, it improved decision making by providing actionable insights through multimodal data integration. As of today, our system has been used by over 1000 students with SENs in Hong Kong, Singapore, and Canada, demonstrating the real-world impact of AI-driven LA.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1097-1111"},"PeriodicalIF":4.9,"publicationDate":"2025-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11270997","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-18DOI: 10.1109/TLT.2025.3634216
Sikai Wang;Xinyi Luo;Khe Foon Hew
Effective self-regulation is crucial for student success; however, many students struggle with it, especially during online activities, leading to disengagement. Existing methods to boost self-regulated learning (SRL) skills, such as writing self-reflective reports and using prompts in video lectures, lack timely personalized feedback and can be labor intensive. This study implemented the Self-Regulated Learning AI Mentor (SRLMentor), a system designed to support students’ SRL skills, including goal-setting, planning, help seeking, and reflection. SRLMentor comprises three integrated modules to provide timely customized SRL feedback to each student. The first module features an SRL knowledge base that stores user chat records and relevant memory-driven adaptive SRL prompts, addressing a significant limitation of large language models—inability to store new experiences in long-term memory during a dialogue. The second module incorporates a retrieval augmented generation (RAG) component to reduce content hallucinations, ensuring that students receive accurate information. The third module provides in-context learning examples that instruct the AI-based chatbot system to produce relevant SRL responses. We evaluated SRLMentor in an eight-session course with 25 students. Our assessment focused on RAG's performance in terms of factual consistency, answer correctness, and semantic similarity; accuracy of SRLMentor's detection of students’ goals and plans compared to human coders; quality of SRLMentor's feedback; and students’ perceptions of the system's usefulness. The results revealed that RAG enhanced factual, correctness, and semantic accuracy of responses. In addition, SRLMentor's assessments of students’ goals and plans closely matched those of human coders. The cluster analysis revealed that students who engaged more with SRLMentor exhibited greater improvement in SRL skills and course knowledge compared to those who engaged less frequently with the system.
{"title":"Enhancing Self-Regulated Learning Using Generative AI: Development and Evaluation of SRLMentor","authors":"Sikai Wang;Xinyi Luo;Khe Foon Hew","doi":"10.1109/TLT.2025.3634216","DOIUrl":"https://doi.org/10.1109/TLT.2025.3634216","url":null,"abstract":"Effective self-regulation is crucial for student success; however, many students struggle with it, especially during online activities, leading to disengagement. Existing methods to boost self-regulated learning (SRL) skills, such as writing self-reflective reports and using prompts in video lectures, lack timely personalized feedback and can be labor intensive. This study implemented the Self-Regulated Learning AI Mentor (<italic>SRLMentor</i>), a system designed to support students’ SRL skills, including goal-setting, planning, help seeking, and reflection. <italic>SRLMentor</i> comprises three integrated modules to provide timely customized SRL feedback to each student. The first module features an SRL knowledge base that stores user chat records and relevant memory-driven adaptive SRL prompts, addressing a significant limitation of large language models—inability to store new experiences in long-term memory during a dialogue. The second module incorporates a retrieval augmented generation (RAG) component to reduce content hallucinations, ensuring that students receive accurate information. The third module provides in-context learning examples that instruct the AI-based chatbot system to produce relevant SRL responses. We evaluated <italic>SRLMentor</i> in an eight-session course with 25 students. Our assessment focused on RAG's performance in terms of factual consistency, answer correctness, and semantic similarity; accuracy of <italic>SRLMentor's</i> detection of students’ goals and plans compared to human coders; quality of <italic>SRLMentor's</i> feedback; and students’ perceptions of the system's usefulness. The results revealed that RAG enhanced factual, correctness, and semantic accuracy of responses. In addition, <italic>SRLMentor's</i> assessments of students’ goals and plans closely matched those of human coders. The cluster analysis revealed that students who engaged more with <italic>SRLMentor</i> exhibited greater improvement in SRL skills and course knowledge compared to those who engaged less frequently with the system.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1048-1061"},"PeriodicalIF":4.9,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674716","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-14DOI: 10.1109/TLT.2025.3630117
Yao Rong;Kathrin Seßler;Emek Gözlüklü;Enkelejda Kasneci
The use of large language models (LLMs) in mathematical reasoning has become a cornerstone of related research, demonstrating the intelligence of these models and enabling potential practical applications through their advanced performance, such as in educational settings. Despite the variety of datasets and in-context learning algorithms designed to improve the ability of LLMs to automate mathematical problem solving, the lack of comprehensive benchmarking across different datasets makes it difficult to determine which in-context algorithms are effective, efficient, and suitable for specific educational applications. In this project, we present a benchmark that fairly compares seven state-of-the-art in-context learning algorithms for mathematical problem solving across five widely used mathematical datasets on four powerful foundation models. Beyond accuracy, we explore the tradeoff between efficiency and performance, highlighting the practical applications of LLMs for mathematical reasoning. Our results indicate that larger foundation models, such as GPT-4o and LLaMA 3-70B, can solve mathematical reasoning independently from the concrete prompting strategy, while for smaller models, the in-context learning approach significantly influences the performance. Moreover, the optimal prompt depends on the chosen foundation model. We open source our benchmark code to support the integration of additional models in future research.
{"title":"Benchmarking In-Context Learning Strategies of Large Language Models for Math Reasoning Tasks","authors":"Yao Rong;Kathrin Seßler;Emek Gözlüklü;Enkelejda Kasneci","doi":"10.1109/TLT.2025.3630117","DOIUrl":"https://doi.org/10.1109/TLT.2025.3630117","url":null,"abstract":"The use of large language models (LLMs) in mathematical reasoning has become a cornerstone of related research, demonstrating the intelligence of these models and enabling potential practical applications through their advanced performance, such as in educational settings. Despite the variety of datasets and in-context learning algorithms designed to improve the ability of LLMs to automate mathematical problem solving, the lack of comprehensive benchmarking across different datasets makes it difficult to determine which in-context algorithms are effective, efficient, and suitable for specific educational applications. In this project, we present a benchmark that fairly compares seven state-of-the-art in-context learning algorithms for mathematical problem solving across five widely used mathematical datasets on four powerful foundation models. Beyond accuracy, we explore the tradeoff between efficiency and performance, highlighting the practical applications of LLMs for mathematical reasoning. Our results indicate that larger foundation models, such as GPT-4o and LLaMA 3-70B, can solve mathematical reasoning independently from the concrete prompting strategy, while for smaller models, the in-context learning approach significantly influences the performance. Moreover, the optimal prompt depends on the chosen foundation model. We open source our benchmark code to support the integration of additional models in future research.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1074-1082"},"PeriodicalIF":4.9,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11249497","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778408","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-10DOI: 10.1109/TLT.2025.3630658
César Domínguez;Arturo Jaime;Beatriz Pérez;Ángel Luis Rubio;María Antonia Zapata
Student generated questions (SGQ) is a constructive educational strategy in which students elaborate their own questions about the contents being learned. Research on this learning method has been focused on academic results, but other important aspects have been overlooked. In this work, we present an innovative, online, and collaborative software application to specifically support the SGQ strategy. The traces left on the tool by 221 students organized in teams are analyzed using process mining, in order to obtain insights from the learning process and the collaboration among students. Using a new feature model to identify the key characteristics of the SGQ strategy, we focus on the quality of the generated questions, the collaborative processes among students during question generation, and the alignment of students’ behavior with the instructors’ plan. In addition, the study is enriched by the influence of some cross-cutting parameters: type of subject, academic level of students, number of questions developed by each student, and availability of the questions–answers for self-study. The results obtained suggest that students were able to formulate good-quality questions and were well-suited to the planned task; however a competitive effect between teams was detected. Furthermore, we found that neither the type of subject nor the academic level of the undergraduates significantly influenced the process. In contrast, the volume and perceived usefulness of the questions did influence the studied characteristics, with lower workload and higher usefulness positively impacting the process. The results obtained thanks to the use of educational process mining on an SGQ learning tool offer valuable guidance for future proposals of this successful learning strategy.
{"title":"Process Mining Insights From a Student-Generated Questions Tool: Lower Workload and Higher Perceived Usefulness Improve the Learning Process","authors":"César Domínguez;Arturo Jaime;Beatriz Pérez;Ángel Luis Rubio;María Antonia Zapata","doi":"10.1109/TLT.2025.3630658","DOIUrl":"https://doi.org/10.1109/TLT.2025.3630658","url":null,"abstract":"Student generated questions (SGQ) is a constructive educational strategy in which students elaborate their own questions about the contents being learned. Research on this learning method has been focused on academic results, but other important aspects have been overlooked. In this work, we present an innovative, online, and collaborative software application to specifically support the SGQ strategy. The traces left on the tool by 221 students organized in teams are analyzed using process mining, in order to obtain insights from the learning process and the collaboration among students. Using a new <italic>feature model</i> to identify the key characteristics of the SGQ strategy, we focus on the quality of the generated questions, the collaborative processes among students during question generation, and the alignment of students’ behavior with the instructors’ plan. In addition, the study is enriched by the influence of some cross-cutting parameters: type of subject, academic level of students, number of questions developed by each student, and availability of the questions–answers for self-study. The results obtained suggest that students were able to formulate good-quality questions and were well-suited to the planned task; however a competitive effect between teams was detected. Furthermore, we found that neither the type of subject nor the academic level of the undergraduates significantly influenced the process. In contrast, the volume and perceived usefulness of the questions did influence the studied characteristics, with lower workload and higher usefulness positively impacting the process. The results obtained thanks to the use of educational process mining on an SGQ learning tool offer valuable guidance for future proposals of this successful learning strategy.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1083-1096"},"PeriodicalIF":4.9,"publicationDate":"2025-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11235992","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145778409","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-06DOI: 10.1109/TLT.2025.3629584
Pablo García-Zarza;Guillermo Vega-Gorgojo;Miguel L. Bote-Lorenzo;Vanesa Gallego-Lema;Juan I. Asensio-Pérez;Eduardo Gómez-Sánchez
Ubiquitous learning (u-learning) leverages educational technologies to help students learn anywhere and anytime across multiple physical and virtual spaces. However, u-learning applications face a challenging tradeoff: Should they provide a predefined set of u-learning resources, thus saving time for teachers, but limiting their applicability to a wider range of u-learning situations? Or should they allow teachers to create their own u-learning resources, improving flexibility, but requiring a nonnegligible effort from teachers that typically ends up in learning resources that cannot be reused by other teachers or by other u-learning applications? Cultural Heritage Educational Semantic Tool (CHEST), the application presented in this article, addresses this tradeoff proposing the use (and reuse) of Linked Open Data (LOD) to support teachers in designing u-learning situations in the cultural heritage domain. CHEST hides the complexity of LOD to teachers, thus reducing the effort in creating u-learning situations, while, at the same time, taking advantage of its reusable nature. CHEST allows teachers to create and reuse three types of learning resources in the form of LOD: spatial things, learning tasks, and itineraries (which group the other two types of resources). This article elicits the requirements considered for the development of CHEST, describes its architecture, and presents the results of an evaluation study carried out with a CHEST prototype in the context of a university course involving two teachers and 14 students. The evaluation examines how CHEST supports teachers in the creation and reuse of u-learning resources based on LOD, paying attention to the balance between flexibility and required effort, while it also showcases how CHEST supports the enactment of u-learning situations in an authentic educational context. The study provides valuable insights into the applicability and effectiveness of CHEST within a specific educational context.
{"title":"CHEST: An Application to Support Teachers in the Use of Linked Open Data for Ubiquitous Learning","authors":"Pablo García-Zarza;Guillermo Vega-Gorgojo;Miguel L. Bote-Lorenzo;Vanesa Gallego-Lema;Juan I. Asensio-Pérez;Eduardo Gómez-Sánchez","doi":"10.1109/TLT.2025.3629584","DOIUrl":"https://doi.org/10.1109/TLT.2025.3629584","url":null,"abstract":"Ubiquitous learning (u-learning) leverages educational technologies to help students learn anywhere and anytime across multiple physical and virtual spaces. However, u-learning applications face a challenging tradeoff: Should they provide a predefined set of u-learning resources, thus saving time for teachers, but limiting their applicability to a wider range of u-learning situations? Or should they allow teachers to create their own u-learning resources, improving flexibility, but requiring a nonnegligible effort from teachers that typically ends up in learning resources that cannot be reused by other teachers or by other u-learning applications? Cultural Heritage Educational Semantic Tool (CHEST), the application presented in this article, addresses this tradeoff proposing the use (and reuse) of Linked Open Data (LOD) to support teachers in designing u-learning situations in the cultural heritage domain. CHEST hides the complexity of LOD to teachers, thus reducing the effort in creating u-learning situations, while, at the same time, taking advantage of its reusable nature. CHEST allows teachers to create and reuse three types of learning resources in the form of LOD: spatial things, learning tasks, and itineraries (which group the other two types of resources). This article elicits the requirements considered for the development of CHEST, describes its architecture, and presents the results of an evaluation study carried out with a CHEST prototype in the context of a university course involving two teachers and 14 students. The evaluation examines how CHEST supports teachers in the creation and reuse of u-learning resources based on LOD, paying attention to the balance between flexibility and required effort, while it also showcases how CHEST supports the enactment of u-learning situations in an authentic educational context. The study provides valuable insights into the applicability and effectiveness of CHEST within a specific educational context.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1021-1035"},"PeriodicalIF":4.9,"publicationDate":"2025-11-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145612131","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-24DOI: 10.1109/TLT.2025.3625146
Penghe Chen;Zhilin Fan;Yu Lu
Students’ problem behaviors, which deviate from established school norms, can undermine both their well-being and academic achievement. Effectively addressing such behaviors often requires interdisciplinary expertise that extends beyond the typical scope of teachers’ professional knowledge. To bridge this gap, we propose KnowSTU, an intelligent assistant powered by large language models (LLMs) to support the diagnosis of students’ problem behaviors. KnowSTU integrates domain-specific LLM fine-tuning with retrieval-augmented generation (RAG) to enhance diagnostic accuracy and deliver actionable educational strategies. Specifically, we construct a theoretical framework for problem behavior diagnosis to guide system design and dataset development, compile a multiturn dialogue dataset of real-world annotated cases, fine-tune a domain-adapted LLM using the quantized low-rank adaptation method to enable context-aware diagnostic conversations, and incorporate an RAG framework to improve contextual relevance and response specificity. Experimental results demonstrate that KnowSTU consistently outperforms baseline LLMs across multiple technical and educational evaluation metrics, confirming its diagnostic effectiveness and practical utility. Moreover, findings from a teacher study reveal strong user acceptance, underscoring the system’s feasibility and educational value in supporting problem behavior diagnosis in classroom contexts.
{"title":"KnowSTU: Diagnosing Students’ Problem Behaviors Using Fine-Tuned LLM and RAG","authors":"Penghe Chen;Zhilin Fan;Yu Lu","doi":"10.1109/TLT.2025.3625146","DOIUrl":"https://doi.org/10.1109/TLT.2025.3625146","url":null,"abstract":"Students’ problem behaviors, which deviate from established school norms, can undermine both their well-being and academic achievement. Effectively addressing such behaviors often requires interdisciplinary expertise that extends beyond the typical scope of teachers’ professional knowledge. To bridge this gap, we propose KnowSTU, an intelligent assistant powered by large language models (LLMs) to support the diagnosis of students’ problem behaviors. KnowSTU integrates domain-specific LLM fine-tuning with retrieval-augmented generation (RAG) to enhance diagnostic accuracy and deliver actionable educational strategies. Specifically, we construct a theoretical framework for problem behavior diagnosis to guide system design and dataset development, compile a multiturn dialogue dataset of real-world annotated cases, fine-tune a domain-adapted LLM using the quantized low-rank adaptation method to enable context-aware diagnostic conversations, and incorporate an RAG framework to improve contextual relevance and response specificity. Experimental results demonstrate that KnowSTU consistently outperforms baseline LLMs across multiple technical and educational evaluation metrics, confirming its diagnostic effectiveness and practical utility. Moreover, findings from a teacher study reveal strong user acceptance, underscoring the system’s feasibility and educational value in supporting problem behavior diagnosis in classroom contexts.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"962-975"},"PeriodicalIF":4.9,"publicationDate":"2025-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145510180","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-20DOI: 10.1109/TLT.2025.3624024
Ricardo Martin Fernandez;Felix Garcia Loro;Elio Sancristóbal;Miguel Rodriguez-Artacho;Hamadou Saliah-Hassane;Manuel Castro
The adoption of standards plays a decisive role in achieving interoperability in educational technology. Remote laboratories have already proven their value in engineering education, but the variety of isolated solutions limits the consolidation of a shared framework. This article explores how pocket labs—decentralized low-cost remote laboratories—can be aligned with the IEEE 1876-2019 standard and outlines a roadmap for their integration into the Lab as a Service and Learning Object layers defined by the standard. Using a case study based on accessible hardware, we show that it is possible to create compliant laboratories while remaining adaptable to contexts where traditional infrastructures are not available. Beyond the technical alignment, the study examines how pocket labs can enrich learning by supporting experimental practice and teamwork by means of a comparison with initiatives such as LabsLand and WebLab-Deusto and illustrates both shared challenges and specific advantages of this approach, and how they tackle the compliance requirements of the standard.
{"title":"Building Compliant Pocket Labs With IEEE STD 1876-2019: A Step Forward","authors":"Ricardo Martin Fernandez;Felix Garcia Loro;Elio Sancristóbal;Miguel Rodriguez-Artacho;Hamadou Saliah-Hassane;Manuel Castro","doi":"10.1109/TLT.2025.3624024","DOIUrl":"https://doi.org/10.1109/TLT.2025.3624024","url":null,"abstract":"The adoption of standards plays a decisive role in achieving interoperability in educational technology. Remote laboratories have already proven their value in engineering education, but the variety of isolated solutions limits the consolidation of a shared framework. This article explores how pocket labs—decentralized low-cost remote laboratories—can be aligned with the IEEE 1876-2019 standard and outlines a roadmap for their integration into the Lab as a Service and Learning Object layers defined by the standard. Using a case study based on accessible hardware, we show that it is possible to create compliant laboratories while remaining adaptable to contexts where traditional infrastructures are not available. Beyond the technical alignment, the study examines how pocket labs can enrich learning by supporting experimental practice and teamwork by means of a comparison with initiatives such as LabsLand and WebLab-Deusto and illustrates both shared challenges and specific advantages of this approach, and how they tackle the compliance requirements of the standard.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"976-988"},"PeriodicalIF":4.9,"publicationDate":"2025-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145510177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-20DOI: 10.1109/TLT.2025.3624050
Lihang Guan;Yue Zhang;Mingyue Michelle Gu
Generative artificial intelligence (GenAI) has demonstrated its benefits for students learning English as a foreign language (EFL). However, this does not mean that teachers are obsolete when artificial intelligence in education (AIED) is employed. Grounded in self-determination theory and the holistic integration approach, this study explored the significance of human agency in AIED to measure three learning outcomes: students’ intrinsic motivation, classroom anxiety, and willingness to communicate in EFL classes. Its findings suggest that students who received human-centered teacher–student–GenAI collaboration developed better in all three areas than those who only had student–GenAI interactions. Moreover, the holistic integration approach promoted teacher immediacy and teacher–student rapport that supported students’ development. In a climate where an instrumentalist view of education is prevalent and digital devices are commonly banned, this study interpreted the changing roles of EFL teachers under AIED systems, suggesting ways that students can benefit from both technological advancements and human agency in education.
{"title":"Future Changes in Teachers’ Professional Roles Under the Impact of Artificial Intelligence: A Study in English as a Foreign Language Education","authors":"Lihang Guan;Yue Zhang;Mingyue Michelle Gu","doi":"10.1109/TLT.2025.3624050","DOIUrl":"https://doi.org/10.1109/TLT.2025.3624050","url":null,"abstract":"Generative artificial intelligence (GenAI) has demonstrated its benefits for students learning English as a foreign language (EFL). However, this does not mean that teachers are obsolete when artificial intelligence in education (AIED) is employed. Grounded in self-determination theory and the holistic integration approach, this study explored the significance of human agency in AIED to measure three learning outcomes: students’ intrinsic motivation, classroom anxiety, and willingness to communicate in EFL classes. Its findings suggest that students who received human-centered teacher–student–GenAI collaboration developed better in all three areas than those who only had student–GenAI interactions. Moreover, the holistic integration approach promoted teacher immediacy and teacher–student rapport that supported students’ development. In a climate where an instrumentalist view of education is prevalent and digital devices are commonly banned, this study interpreted the changing roles of EFL teachers under AIED systems, suggesting ways that students can benefit from both technological advancements and human agency in education.","PeriodicalId":49191,"journal":{"name":"IEEE Transactions on Learning Technologies","volume":"18 ","pages":"1062-1073"},"PeriodicalIF":4.9,"publicationDate":"2025-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：