Hsin-Yu Lee
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Hsin-Yu Lee has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Yueh-Min Huang
National Cheng Kung University
30 shared publications - Ting-Ting Wu
National Yunlin University of Science and Technology
25 shared publications - Chia-Ju Lin
National Cheng Kung University
19 shared publications - Wei-Sheng Wang
National Yunlin University of Science and Technology
17 shared publications - Pei-Hua Chen
National Cheng Kung University
10 shared publications - Pin-Hui Li
Department of Engineering Science National Cheng Kung University Tainan Taiwan
5 shared publications
Latest Publications
- 10 September 2026
- Published by SAGE Publications in Journal of Educational Computing Research
Differentiated Formative Feedback Through Parameterized ChatGPT: Effects on Knowledge Construction, Engagement, and Higher-Order Thinking in STEM Education
STEM education increasingly emphasizes interdisciplinary integration and self-directed learning, yet students with different achievement levels often face unequal challenges when navigating complex cross-domain content. High-achieving students may lack sufficient cognitive stimulation, whereas low-achieving students may struggle with overwhelming complexity. Standard ChatGPT feedback lacks adaptability to these diverse learner capabilities, leaving some students unchallenged and others overloaded. Addressing these disparities requires timely and differentiated instructional support. This study proposes AdjustGPT, a system that calibrates ChatGPT’s generative parameters (temperature, max_tokens, frequency_penalty, presence_penalty, and top_p) according to learners’ achievement levels to deliver differentiated formative feedback. Grounded in formative feedback theory and cognitive load principles, AdjustGPT provides high-achieving students with exploratory, multi-perspective feedback that fosters critical thinking and innovation, while offering low-achieving students clear, structured guidance that reduces cognitive load and strengthens foundational understanding. A quasi-experimental study with 62 university students in a STEM course showed that, within this exploratory sample, AdjustGPT was associated with significant gains in knowledge construction, cognitive and emotional engagement, and higher-order thinking outcomes, including critical thinking and problem-solving. These findings provide initial evidence that parameterized ChatGPT systems can support adaptive formative feedback and differentiated AI-assisted teaching in STEM education. Larger-scale studies are needed to establish robustness and generalizability.
- 10 September 2026
- Published by SAGE Publications in Journal of Educational Computing Research
Differentiated Formative Feedback Through Parameterized ChatGPT: Effects on Knowledge Construction, Engagement, and Higher-Order Thinking in STEM Education
STEM education increasingly emphasizes interdisciplinary integration and self-directed learning, yet students with different achievement levels often face unequal challenges when navigating complex cross-domain content. High-achieving students may lack sufficient cognitive stimulation, whereas low-achieving students may struggle with overwhelming complexity. Standard ChatGPT feedback lacks adaptability to these diverse learner capabilities, leaving some students unchallenged and others overloaded. Addressing these disparities requires timely and differentiated instructional support. This study proposes AdjustGPT, a system that calibrates ChatGPT’s generative parameters (temperature, max_tokens, frequency_penalty, presence_penalty, and top_p) according to learners’ achievement levels to deliver differentiated formative feedback. Grounded in formative feedback theory and cognitive load principles, AdjustGPT provides high-achieving students with exploratory, multi-perspective feedback that fosters critical thinking and innovation, while offering low-achieving students clear, structured guidance that reduces cognitive load and strengthens foundational understanding. A quasi-experimental study with 62 university students in a STEM course showed that, within this exploratory sample, AdjustGPT was associated with significant gains in knowledge construction, cognitive and emotional engagement, and higher-order thinking outcomes, including critical thinking and problem-solving. These findings provide initial evidence that parameterized ChatGPT systems can support adaptive formative feedback and differentiated AI-assisted teaching in STEM education. Larger-scale studies are needed to establish robustness and generalizability.
- Open Access
- 16 July 2026
- Published by Springer Nature in International Journal of STEM Education
- Vol. 13 (1) , 33
Unpacking the heterogeneity of pre-service teachers’ ChatGPT acceptance: a latent profile analysis across STEM and non-STEM disciplines
- 16 July 2026
- Published by Springer Nature in International Journal of STEM Education
- Vol. 13 (1) , 33
Open AccessUnpacking the heterogeneity of pre-service teachers’ ChatGPT acceptance: a latent profile analysis across STEM and non-STEM disciplines
- 22 February 2026
- Published by Wiley in European Journal of Education
- Vol. 61 (1)
Enhancing Flipped Learning With Convergent Feedback: The Impact of GPTutor on Knowledge Construction, Motivation and Engagement in Higher Education
While flipped learning promotes active learning in higher education, students often struggle with self‐regulation and motivation when facing complex content. Traditional search engines provide divergent feedback that can increase cognitive load and impede learning effectiveness. This study examined the impact of GPTutor, an AI‐powered tool integrating ChatGPT and Apple's Shortcuts to provide convergent feedback, on knowledge construction, motivation and engagement. Using a quasi‐experimental design, 72 Taiwanese educational technology students were divided into control (Google search) and experimental (GPTutor) groups. ANCOVA and t ‐tests revealed that GPTutor significantly improved post‐test scores ( F = 7.99, p < 0.01), intrinsic motivation ( t = 2.993, p < 0.01) and reduced amotivation ( t = −2.307, p < 0.05). The experimental group also demonstrated higher cognitive, behavioural and emotional engagement (all p < 0.01). The findings suggest that AI‐powered convergent feedback can enhance flipped learning effectiveness in higher education, particularly for complex interdisciplinary content.
- 22 February 2026
- Published by Wiley in European Journal of Education
- Vol. 61 (1)
Enhancing Flipped Learning With Convergent Feedback: The Impact of GPTutor on Knowledge Construction, Motivation and Engagement in Higher Education
While flipped learning promotes active learning in higher education, students often struggle with self‐regulation and motivation when facing complex content. Traditional search engines provide divergent feedback that can increase cognitive load and impede learning effectiveness. This study examined the impact of GPTutor, an AI‐powered tool integrating ChatGPT and Apple's Shortcuts to provide convergent feedback, on knowledge construction, motivation and engagement. Using a quasi‐experimental design, 72 Taiwanese educational technology students were divided into control (Google search) and experimental (GPTutor) groups. ANCOVA and t ‐tests revealed that GPTutor significantly improved post‐test scores ( F = 7.99, p < 0.01), intrinsic motivation ( t = 2.993, p < 0.01) and reduced amotivation ( t = −2.307, p < 0.05). The experimental group also demonstrated higher cognitive, behavioural and emotional engagement (all p < 0.01). The findings suggest that AI‐powered convergent feedback can enhance flipped learning effectiveness in higher education, particularly for complex interdisciplinary content.
- 21 February 2026
- Published by SAGE Publications in Journal of Educational Computing Research
- Vol. 64 (5) , 1254-1293
From Immediate Answers to Structured Inquiry: A Socratic Method-Enhanced ChatGPT System for Blended Learning Discussions
Artificial Intelligence integration with mobile instant messaging (MIM) offers promising support for blended learning discussions, yet current implementations often favor immediacy over pedagogical structure. Despite separate studies on Socratic methods, ChatGPT applications, and MIM platforms, their systematic integration remains unexplored. This study introduces SCMIM (Socratic-based ChatGPT Mobile Instant Messaging) and compares its impact with standard ChatGPT-MIM and traditional MIM through a 12-week randomized trial with 72 master’s students. Using mixed-methods analysis of engagement questionnaires, discussion logs, and interviews, we examined behavioral, cognitive, and emotional engagement alongside technology acceptance perceptions. Results showed ChatGPT-MIM enhanced behavioral participation and emotional comfort through immediate support, while SCMIM more effectively fostered cognitive engagement, particularly analytical thinking and knowledge construction. Despite initial adaptation challenges, SCMIM users developed appreciation for systematic questioning, reporting high perceived usefulness and comparable intention to use. These findings demonstrate that effective AI integration requires balancing immediate support with structured pedagogical guidance to promote both engagement and deeper learning outcomes in educational discussions.
- 21 February 2026
- Published by SAGE Publications in Journal of Educational Computing Research
- Vol. 64 (5) , 1254-1293
From Immediate Answers to Structured Inquiry: A Socratic Method-Enhanced ChatGPT System for Blended Learning Discussions
Artificial Intelligence integration with mobile instant messaging (MIM) offers promising support for blended learning discussions, yet current implementations often favor immediacy over pedagogical structure. Despite separate studies on Socratic methods, ChatGPT applications, and MIM platforms, their systematic integration remains unexplored. This study introduces SCMIM (Socratic-based ChatGPT Mobile Instant Messaging) and compares its impact with standard ChatGPT-MIM and traditional MIM through a 12-week randomized trial with 72 master’s students. Using mixed-methods analysis of engagement questionnaires, discussion logs, and interviews, we examined behavioral, cognitive, and emotional engagement alongside technology acceptance perceptions. Results showed ChatGPT-MIM enhanced behavioral participation and emotional comfort through immediate support, while SCMIM more effectively fostered cognitive engagement, particularly analytical thinking and knowledge construction. Despite initial adaptation challenges, SCMIM users developed appreciation for systematic questioning, reporting high perceived usefulness and comparable intention to use. These findings demonstrate that effective AI integration requires balancing immediate support with structured pedagogical guidance to promote both engagement and deeper learning outcomes in educational discussions.
- 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 230-236
Mitigating Artificial Intelligence Hallucinations in Education: A Comparative Study of Retrieval-Augmented Generation (RAG) and Large Language Models
The integration of Large Language Models (LLMs) into educational technologies promises to revolutionize personalized learning. However, their propensity for “hallucination,” the generation of factually incorrect or nonsensical information, poses a significant risk to knowledge integrity and student trust. To address this critical challenge, this paper investigates Retrieval-Augmented Generation (RAG), an architectural approach that mitigates hallucinations by grounding the model's responses in factual data. Before generating an answer, the RAG framework retrieves relevant information from a verified knowledge base, ensuring the output is contextually accurate. We conducted a comparative study between a standard LLM and a RAG-powered system whose knowledge base was populated with specific curriculum materials. The generated answers to domain-specific questions were assessed through two distinct lenses: a quantitative evaluation of factual accuracy by subject matter experts and a qualitative analysis of student perceptions gathered through semi-structured interviews. Our preliminary findings indicate that the RAG model demonstrates a marked improvement in factual accuracy and a significant reduction in hallucinatory content. Furthermore, students perceived the RAG-generated responses as more trustworthy and useful for their learning. This research provides empirical evidence for the necessity of adopting RAG in educational AI, offering a pathway toward developing more reliable and effective digital learning tools.
- 18 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 230-236
Mitigating Artificial Intelligence Hallucinations in Education: A Comparative Study of Retrieval-Augmented Generation (RAG) and Large Language Models
The integration of Large Language Models (LLMs) into educational technologies promises to revolutionize personalized learning. However, their propensity for “hallucination,” the generation of factually incorrect or nonsensical information, poses a significant risk to knowledge integrity and student trust. To address this critical challenge, this paper investigates Retrieval-Augmented Generation (RAG), an architectural approach that mitigates hallucinations by grounding the model's responses in factual data. Before generating an answer, the RAG framework retrieves relevant information from a verified knowledge base, ensuring the output is contextually accurate. We conducted a comparative study between a standard LLM and a RAG-powered system whose knowledge base was populated with specific curriculum materials. The generated answers to domain-specific questions were assessed through two distinct lenses: a quantitative evaluation of factual accuracy by subject matter experts and a qualitative analysis of student perceptions gathered through semi-structured interviews. Our preliminary findings indicate that the RAG model demonstrates a marked improvement in factual accuracy and a significant reduction in hallucinatory content. Furthermore, students perceived the RAG-generated responses as more trustworthy and useful for their learning. This research provides empirical evidence for the necessity of adopting RAG in educational AI, offering a pathway toward developing more reliable and effective digital learning tools.
- Top Cited Papers
- 18 September 2025
- Published by Wiley in European Journal of Education
- Vol. 60 (4)
ChatGPT in Education: A Systematic Review of Current Landscape, Limitations and Future Directions Through General System Theory Lens
The rapid proliferation of ChatGPT has fuelled expectations of transformative change in education; yet most systematic reviews remain confined to single disciplines or educational levels. Adopting general system theory (GST) and the knowledge‐skills‐attitudes (KSA) framework, this review offers a more comprehensive perspective on ChatGPT's educational impact. Following PRISMA guidelines, we searched major academic databases for studies published between November 2022 and March 2024, screened them against predefined criteria, and ultimately analysed 92 eligible papers through qualitative content analysis. Findings were mapped onto the five GST elements—subject, information, medium, environment and technology. Research to date is markedly concentrated in higher education (69.7%), with far less work in primary (3.4%) and secondary (13.5%) settings. Nearly half of the studies target STEM subjects (45.6%). The most frequently reported applications are timely feedback (55 occurrences), content generation (20) and personalised learning (15). Positive effects are most evident for knowledge construction (63 occurrences) and skills development (59), whereas attitudinal outcomes receive comparatively little attention (20). Overall, the evidence points to promising—but uneven—educational benefits. Prevailing limitations include small‐ or medium‐scale research designs and the under‐representation of primary and secondary contexts. Future work should adopt larger samples, give greater weight to attitudinal dimensions, and strive for a more balanced coverage across all educational levels to realise ChatGPT's full potential in teaching and learning.
- 18 September 2025
- Published by Wiley in European Journal of Education
- Vol. 60 (4)
Top Cited PapersChatGPT in Education: A Systematic Review of Current Landscape, Limitations and Future Directions Through General System Theory Lens
The rapid proliferation of ChatGPT has fuelled expectations of transformative change in education; yet most systematic reviews remain confined to single disciplines or educational levels. Adopting general system theory (GST) and the knowledge‐skills‐attitudes (KSA) framework, this review offers a more comprehensive perspective on ChatGPT's educational impact. Following PRISMA guidelines, we searched major academic databases for studies published between November 2022 and March 2024, screened them against predefined criteria, and ultimately analysed 92 eligible papers through qualitative content analysis. Findings were mapped onto the five GST elements—subject, information, medium, environment and technology. Research to date is markedly concentrated in higher education (69.7%), with far less work in primary (3.4%) and secondary (13.5%) settings. Nearly half of the studies target STEM subjects (45.6%). The most frequently reported applications are timely feedback (55 occurrences), content generation (20) and personalised learning (15). Positive effects are most evident for knowledge construction (63 occurrences) and skills development (59), whereas attitudinal outcomes receive comparatively little attention (20). Overall, the evidence points to promising—but uneven—educational benefits. Prevailing limitations include small‐ or medium‐scale research designs and the under‐representation of primary and secondary contexts. Future work should adopt larger samples, give greater weight to attitudinal dimensions, and strive for a more balanced coverage across all educational levels to realise ChatGPT's full potential in teaching and learning.
- 8 September 2025
- Published by Springer Nature in Education and Information Technologies
- Vol. 30 (17) , 24815-24854
Leveraging ChatGPT for personalized reflective learning in programming education: effects on self-efficacy, higher-order thinking, and project implementation skills
- 8 September 2025
- Published by Springer Nature in Education and Information Technologies
- Vol. 30 (17) , 24815-24854
Leveraging ChatGPT for personalized reflective learning in programming education: effects on self-efficacy, higher-order thinking, and project implementation skills
- 3 September 2025
- Published by Wiley in Journal of Computer Assisted Learning
- Vol. 41 (5)
Integrating Computational Thinking Diagnostic Mechanism and Reflective Learning: An Innovative Approach to Enhance Learning Outcomes in Introductory Programming
Background: Conventional reflective learning methodologies in programming education often lack structured guidance and individualised feedback, limiting their pedagogical effectiveness. Whilst computational thinking (CT) offers a systematic problem‐solving framework with decomposition, pattern recognition, abstraction, and algorithm design, its potential application as a diagnostic instrument for reflection remains insufficiently explored within programming education.Objectives: This study aims to develop and evaluate a CT‐based diagnostic reflective report system as a technological intervention to facilitate structured reflective learning in programming education. Furthermore, it investigates the impact of this system on knowledge construction, higher‐order thinking skills (HOTS), and project performance within an introductory Python programming course.Methods: The study employed a quasi‐experimental design spanning two academic semesters, involving 82 undergraduate students randomly assigned to experimental (n = 42) and control (n = 40) groups. The experimental group utilised weekly CT‐based diagnostic reflective reports, whilst the control group engaged in traditional reflective practises. The curriculum integrated Python programming with Raspberry Pi embedded systems. Assessment measures included pre‐ and post‐tests for knowledge construction, a validated questionnaire for HOTS evaluation, and the Creative Product Analysis Matrix (CPAM) for project performance assessment.Results and Conclusions: Implementation of the CT‐based diagnostic reflective report system demonstrated statistically significant improvements in knowledge construction, critical thinking, and problem‐solving skills compared to traditional approaches. Project performance metrics, including valuable, logical, useful, understandable, and well‐crafted, showed marked enhancement. However, no significant impact was observed regarding creativity. These findings substantiate the efficacy of integrating CT diagnostic mechanisms with reflective learning practises.
- 3 September 2025
- Published by Wiley in Journal of Computer Assisted Learning
- Vol. 41 (5)
Integrating Computational Thinking Diagnostic Mechanism and Reflective Learning: An Innovative Approach to Enhance Learning Outcomes in Introductory Programming
Background: Conventional reflective learning methodologies in programming education often lack structured guidance and individualised feedback, limiting their pedagogical effectiveness. Whilst computational thinking (CT) offers a systematic problem‐solving framework with decomposition, pattern recognition, abstraction, and algorithm design, its potential application as a diagnostic instrument for reflection remains insufficiently explored within programming education.Objectives: This study aims to develop and evaluate a CT‐based diagnostic reflective report system as a technological intervention to facilitate structured reflective learning in programming education. Furthermore, it investigates the impact of this system on knowledge construction, higher‐order thinking skills (HOTS), and project performance within an introductory Python programming course.Methods: The study employed a quasi‐experimental design spanning two academic semesters, involving 82 undergraduate students randomly assigned to experimental (n = 42) and control (n = 40) groups. The experimental group utilised weekly CT‐based diagnostic reflective reports, whilst the control group engaged in traditional reflective practises. The curriculum integrated Python programming with Raspberry Pi embedded systems. Assessment measures included pre‐ and post‐tests for knowledge construction, a validated questionnaire for HOTS evaluation, and the Creative Product Analysis Matrix (CPAM) for project performance assessment.Results and Conclusions: Implementation of the CT‐based diagnostic reflective report system demonstrated statistically significant improvements in knowledge construction, critical thinking, and problem‐solving skills compared to traditional approaches. Project performance metrics, including valuable, logical, useful, understandable, and well‐crafted, showed marked enhancement. However, no significant impact was observed regarding creativity. These findings substantiate the efficacy of integrating CT diagnostic mechanisms with reflective learning practises.