Computers and Education Artificial Intelligence最新文献

{"title":"Factors influencing university students’ intention to use and reliance on generative artificial intelligence: An extended technology acceptance model with critical use","authors":"Trang H. Nguyen ,&nbsp;Long T. Truong ,&nbsp;Nhu H.T. Nguyen","doi":"10.1016/j.caeai.2026.100618","DOIUrl":"10.1016/j.caeai.2026.100618","url":null,"abstract":"<div><div>In light of the growing popularity, accessibility, and utility of Generative Artificial Intelligence (GenAI) tools, their impact on learning can be either supportive or detrimental, depending on how students use them. This calls for an examination of students’ intention to use and reliance on GenAI to inform timely interventions within higher education to ensure favourable learning outcomes. This study investigates factors influencing intentions to use and reliance on GenAI among Engineering and Computer Science/Information Technology (CS/IT) students. Drawing on an extended Technology Acceptance Model (TAM), the study integrates the construct of critical use and conceptualises reliance across four functional domains relevant to engineering and CS/IT education. Data from an anonymous survey collected at an Australian university (n = 126) are analysed using the Partial Least Squares Structural Equation Modelling (PLS-SEM) approach. The modelling results show that attitudes towards GenAI and critical use directly and positively influence intention to use GenAI, while perceived ease of use and perceived usefulness have positive indirect effects. The intention to use, in turn, significantly predicts reliance on GenAI across the four domains: understanding, assessment, programming, and engineering projects. Students demonstrate a moderate level of reliance overall, with greater use for understanding-related tasks and limited utilisation for full assessment writing. This study offers insights for higher education institutions aiming to foster ethical and critical use of GenAI for learning.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100618"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148167471","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Retraction notice to “Integrating AI-based adaptive learning into the flipped classroom model to enhance engagement and learning outcomes” [Computers and Education: Artificial Intelligence 8 (2025) 100392]","authors":"Jozsef Katona ,&nbsp;Klara Ida Katonane Gyonyoru","doi":"10.1016/j.caeai.2026.100543","DOIUrl":"10.1016/j.caeai.2026.100543","url":null,"abstract":"","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100543"},"PeriodicalIF":23.4,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148380045","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluating adaptive and generative AI-based feedback and recommendations in a knowledge-graph-integrated programming learning system","authors":"Lalita Na Nongkhai ,&nbsp;Jingyun Wang ,&nbsp;Adam Wynn ,&nbsp;Takahiko Mendori","doi":"10.1016/j.caeai.2025.100526","DOIUrl":"10.1016/j.caeai.2025.100526","url":null,"abstract":"<div><div>This paper introduces the design and development of a framework that integrates a large language model (LLM) with a retrieval-augmented generation (RAG) approach leveraging both a knowledge graph and user interaction history. The framework is incorporated into a previously developed adaptive learning support system to assess learners’ code, generate formative feedback, and recommend exercises. Moreover, this study examines learner preferences across three instructional modes: adaptive, Generative AI (GenAI), and hybrid GenAI–adaptive. An experimental study was conducted to compare the learning performance and perception of the learners, and the effectiveness of these three modes using four key log features derived from 4956 code submissions across all experimental groups. The analysis results show that learners receiving feedback from GenAI modes had significantly more correct code and fewer code submissions missing essential programming logic than those receiving feedback from adaptive mode. In particular, the hybrid GenAI–adaptive mode achieved the highest number of correct submissions and the fewest incorrect or incomplete attempts, outperforming both the adaptive-only and GenAI-only modes. Questionnaire responses further indicated that GenAI-generated feedback was widely perceived as helpful, while all modes were rated positively for ease of use and usefulness. These results suggest that the hybrid GenAI–adaptive mode outperforms the other two modes across all measured log features.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100526"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145718923","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From teachers to chatbots: Scaffolded corrective feedback and student trust in online L2 English classrooms","authors":"Ali Soyoof ,&nbsp;Barry Lee Reynolds ,&nbsp;Ehsan Rassaei ,&nbsp;Chian-Wen Kao ,&nbsp;Xuan Van Ha","doi":"10.1016/j.caeai.2025.100530","DOIUrl":"10.1016/j.caeai.2025.100530","url":null,"abstract":"<div><div>Teacher corrective feedback (TCF) plays a vital role in second language (L2) learning. Recent studies have examined feedback provided by both human teachers and large language models (LLMs). However, little is known about how students' trust differs toward scaffolded corrective feedback (SCF)—that is, feedback that incrementally progresses from indirect to direct during interaction—when it is provided by an LLM such as ChatGPT versus a language teacher. To address this gap, this study compared the effects of SCF, delivered by language teachers and ChatGPT, on L2 learning outcomes and student trust. Using a mixed-methods design, 40 lower-intermediate Iranian learners of English as a foreign language were randomly assigned to two conditions to receive scaffolded CF on English article usage from either a teacher or ChatGPT across four sessions. Learning gains obtained from immediate and delayed post-tests were analyzed using ANOVA and paired-sample <em>t</em>-tests, while semi-structured interviews and feedback interaction logs were examined using thematic analysis. Results showed that students in the teacher-delivered feedback group significantly outperformed those in the ChatGPT-delivered feedback group on both post- and delayed post-tests. Qualitative analyses suggested that this advantage stemmed from higher trust in the teacher, driven by the teacher's personalized emotional and technical support. The findings highlight that while ChatGPT can serve as a feedback tool in L2 instruction, its effectiveness depends on teacher mediation that attends to learners' individual differences and affective needs.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100530"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926740","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Seeking knowledge or efficiency: Profiling students’ AI-use through survey-based latent class analysis","authors":"Mattias W. Hugerth ,&nbsp;Luisa W. Hugerth","doi":"10.1016/j.caeai.2025.100531","DOIUrl":"10.1016/j.caeai.2025.100531","url":null,"abstract":"<div><div>The spread of easily accessible generative AI in the form of chatbots has impacted secondary education, but the effects of this are largely unknown. Previous studies have shown that using chatbots in a learning context can be both harmful or helpful depending on how they are used. While students are undoubtedly utilising this technology, there is scarce data on the extent, intention, or approach to its use, or what drives it.</div><div>The present study builds upon the findings of a previous qualitative study, aiming to investigate and quantify students' use of generative AI for schoolwork. Through a survey sent to multiple upper secondary schools, we collected 1266 responses to analyse upper secondary students' attitudes toward, usage of, support for, and knowledge about generative AI. We present an overview of students’ AI usage and knowledge using descriptive statistics. For further analysis, a Latent Class Analysis was conducted and four distinct response patterns among students identified: AI-positive knowledge-seekers, Cautious AI-adopters, AI-sceptics and Efficiency-seekers. These four classes were then used to explore differences relating to gender, grade, choice of study programme, attitude to knowledge, neuropsychiatric diagnoses and non-native Swedish speaking students.</div><div>We find that students use generative AI for schoolwork primarily as support for the process of doing their schoolwork but also as a shortcut for tasks perceived as meaningless. We find that the identified patterns of attitudes, knowledge and usage exhibit behaviours that are in different ways both promising and worrisome, and that warrant different courses of action in education.</div><div>This research contributes by identifying differences in student behaviour and attitudes towards AI, and points to needs for further research in the diversity of behaviour and the consequences of different use patterns, as well as the need to tailor educational support for different student groups.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100531"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791848","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating artificial intelligence and data envelopment analysis for sustainable efficiency assessment in higher education","authors":"William Villegas-Ch ,&nbsp;Rommel Gutierrez ,&nbsp;Angel Jaramillo-Alcazar ,&nbsp;Alexandra Maldonado-Navarro","doi":"10.1016/j.caeai.2026.100557","DOIUrl":"10.1016/j.caeai.2026.100557","url":null,"abstract":"<div><div>The rapid digitalization of higher education has exposed inconsistencies between increased investments in technological infrastructure, teaching resources, and energy consumption, and the expected improvements in academic performance. This gap highlights the need for predictive and explainable approaches that can evaluate course-level efficiency while incorporating sustainability principles. In this study, Data Envelopment Analysis (DEA) under the output-oriented BCC formulation was applied to fifteen university courses to estimate their relative efficiency based on multiple inputs, including teaching staff, technological investment, digital resource usage, and energy consumption, and outputs such as passing, retention, completion, satisfaction, and learning outcomes. Only three courses achieved full efficiency (<span><math><mi>h</mi><mo>=</mo><mn>1.0</mn></math></span>), whereas the remaining ones showed efficiency gaps ranging from 5.4% to 20.2% relative to the frontier. Energy consumption and learning outcomes emerged as the most influential determinants of efficiency. Artificial intelligence models were then trained under DEA supervision to predict efficiency scores with high fidelity (<span><math><msup><mi>R</mi><mrow><mn>2</mn></mrow></msup><mo>&gt;</mo><mn>0.90</mn></math></span> for Random Forest and Gradient Boosting), replicating DEA behavior and enabling prospective analysis. Finally, indicators aligned with Sustainable Development Goals (SDGs 4 and 12) demonstrated that equitable digital access management reduces dropout rates. At the same time, optimized energy use enhances sustainability metrics by up to twelve points, reinforcing the relevance of AI-driven learning analytics for efficient and sustainable educational management.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100557"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147384879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unleashing human potential: An artificial intelligence competency framework for K–12 education","authors":"Siu Cheung Kong ,&nbsp;Wenxi Hu","doi":"10.1016/j.caeai.2026.100556","DOIUrl":"10.1016/j.caeai.2026.100556","url":null,"abstract":"<div><div>This study explores strategic approaches for integrating artificial intelligence (AI) into K–12 education to prepare learners for the AI era. It addresses two critical questions: 1) What are the key components of AI competency frameworks identified in current research? 2) What framework can guide the effective, responsible integration of AI in K–12 while prioritising human values? We conducted a scoping review of 54 studies, identifying three core student competencies: foundational AI knowledge, practical and cognitive skills of using AI, and ethical awareness. However, existing frameworks place limited emphasis on developing human values. Consequently, we propose a three-component developmental framework—Understanding, Using, and Unleashing—to guide AI integration. The ‘Understanding’ component cultivates a conceptual understanding of AI, while ‘Using’ emphasizes practical and cognitive skills of using AI to enhance learning. Finally, ‘Unleashing’ highlights AI's potential to empower personal growth, and fostering a spiritual self capable of making value-based decisions and do good. This framework aims to prepare learners to contribute meaningfully to future society. Schools must teach students to use AI for empowerment and good, avoiding mere reliance or shortcuts. Through practical activities, education should guide learners to unleash their potential and harness technology for spiritual self-development.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100556"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385021","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring the design of LLM-powered question generation for deaf and hard of hearing learners","authors":"Si Chen ,&nbsp;Shuxu Huffman ,&nbsp;Qingxiaoyang Zhu ,&nbsp;Haotian Su ,&nbsp;Qi Wang ,&nbsp;Raja Kushalnagar","doi":"10.1016/j.caeai.2026.100615","DOIUrl":"10.1016/j.caeai.2026.100615","url":null,"abstract":"<div><div>Deaf and Hard of Hearing (DHH) learners face unique learning challenges, often due to a lack of customized educational materials that address their specific needs. This study explores the potential of Large Language Models (LLMs) to generate personalized quiz questions to enhance the video-based learning experiences of DHH students. We iteratively designed, developed, and evaluated a prototype that combines LLMs to generate questions with the characteristics of the DHH learners in focus. The two unique question generation strategies include Visual Questions, which identify video segments where visual information might be misinterpreted, and Emotion Questions, which highlight moments where previous DHH learners experienced video learning difficulties manifested in emotional responses. Through user studies with undergraduate DHH students, we evaluated the effectiveness of these LLM-powered questions generated to support the learning experience. Our findings indicate that while LLMs offer significant potential for personalized learning, challenges remain in the human-machine interaction caused by inherent accessibility issues between text-based AI prompts and DHH students’ sign-based first/native languages. The study highlights the importance of considering language diversity and culture in the design of LLM-based educational technology.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100615"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148230827","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EvalYaks: Instruction tuning datasets and LoRA fine-tuned models for automated scoring of CEFR B2 speaking assessment transcripts","authors":"Nicy Scaria ,&nbsp;Silvester John Joseph Kennedy ,&nbsp;Thomas Latinovich ,&nbsp;Deepak Subramani","doi":"10.1016/j.caeai.2025.100539","DOIUrl":"10.1016/j.caeai.2025.100539","url":null,"abstract":"<div><div>Relying on human experts to evaluate the Common European Framework of Reference for Languages (CEFR) speaking assessments in an e-learning environment creates scalability challenges, as it limits how quickly and widely assessments can be conducted. We aim to automate the evaluation of CEFR B2 English speaking assessments in e-learning environments from conversation transcripts. First, we evaluate the capability of leading open source and commercial Large Language Models (LLMs) to score a candidate’s performance across various criteria in the CEFR B2 speaking exam in both global and India-specific contexts. Next, we create a new expert-validated, CEFR-aligned synthetic conversational dataset with transcripts that are rated at different assessment scores. In addition, new instruction-tuned datasets are developed from the English Vocabulary Profile (up to CEFR B2 level) and the CEFR-SP WikiAuto datasets. Finally, using these new datasets, we perform parameter efficient instruction tuning of Mistral Instruct 7B v0.2 to develop a family of models called <em>EvalYaks</em>. Four models in this family are for assessing the four sections of the CEFR B2 speaking exam, one for identifying the CEFR level of vocabulary and generating level-specific vocabulary, and another for detecting the CEFR level of text and generating level-specific text. <em>EvalYaks</em> achieved an <em>average acceptable accuracy</em> of 96 %, a degree of variation of 0.35 levels, achieving performance competitive with state-of-the-art frontier models like GPT-4o and Gemini Flash 2.5. Furthermore, a pilot validation on real-world learner transcripts verified the model’s transferability to real-world assessment contexts. This demonstrates that a 7B parameter LLM instruction tuned with high-quality CEFR-aligned assessment data can effectively evaluate and score CEFR B2 English speaking assessments, offering a promising solution for scalable, automated language proficiency evaluation. The methodology is adaptable to other regional contexts and CEFR levels through appropriate data generation and validation protocols.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100539"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926801","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Conversational AI in children's home literacy learning: effectiveness, advantages, challenges, and family perception","authors":"Shuang Quan ,&nbsp;Xintian Tu-Shea ,&nbsp;Yi Ding ,&nbsp;Yao Du ,&nbsp;Qingxiao Zheng ,&nbsp;Laney E. Gerdich","doi":"10.1016/j.caeai.2026.100549","DOIUrl":"10.1016/j.caeai.2026.100549","url":null,"abstract":"<div><div>This study investigates the effectiveness, affordances, limitations, and family perceptions of conversational AI for home literacy learning vs. human. We developed a large language model (LLM)-powered conversational AI system, named Vovo, to teach children vocabulary and co-construct stories using structured literacy pedagogy. The system was tested in home environments over six weeks with 10 families and their children aged 3–7 (<em>M</em> = 5.4). Across 150 learning sessions, Vovo delivered structured literacy instruction as effectively as parents, though children achieved higher learning outcomes when learning with parents. Video analysis revealed Vovo's advantages in pedagogical consistency, language modeling, and verbal socioemotional support, while facing challenges in speech recognition, instructional persistence, nonverbal social cues, and phoneme instruction. Parents perceived Vovo as intelligent, useful, and trustworthy, while expecting a multimodal design to improve engagement. Children perceived Vovo smart and fun but still preferred learning with parents due to emotional bonding. As one of the first studies to embed structured literacy pedagogy into home-based conversational AI system, this research contributes empirical insights into the evolving role of AI in home literacy environments. It also underscores the socially responsive AI design in early education and calls for future design that support parent-child-AI triadic interactions to optimize AI in home literacy learning.</div></div>","PeriodicalId":34469,"journal":{"name":"Computers and Education Artificial Intelligence","volume":"10 ","pages":"Article 100549"},"PeriodicalIF":0.0,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077863","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}