Contribution: This study provides evidence from a structured individual-to-collaborative isomorphic task sequence while explicitly distinguishing group-produced technical performance from student-level psychological associations. Background: Collaborative problem solving can support peer reasoning, but group-produced outcomes should not be conflated with individual learning evidence when repeated observations arise from students and groups. Research Questions: The study examines how collaborative group performance compares with precollaboration individual performance within each group and how intrinsic motivation (IM), professional identity (PI), and knowledge-transfer perception are (kT) associated with perceived academic success (PS). Methodology: A repeated classroom design involved 53 eligible students across eight experimental modules. Seventy verified Group-by-Activity observations were summarized descriptively. Repeated activity occasions were aggregated within stable group clusters before directional inference. The upper-median benchmark was used for the primary directional comparison, with the best-member benchmark used as a diagnostic comparison. Repeated questionnaire responses were aggregated within students. Findings: Relative to the upper-median benchmark, 14 group products were higher, 56 equal, and none lower. Across the ten stable group clusters, four were positive, six were tied, and none were negative; the directional exact test did not reach $\alpha =0.05$ (one-sided $p=0.063$ ; two-sided $p=0.125$ ). Relative to the best-member benchmark, all 70 group products matched the strongest member’s archived category, with complete top-category censoring. IM ( $\rho =0.558$ ), PI ( $\rho =0.422$ ), and KT ( $\rho =0.361$ ) were positively associated with PS. The findings characterize group-produced performance but do not establish a collaboration-specific effect or individual learning gain.
Large language models (LLMs) have emerged as a transformative technological force in modern programming education, offering novel solutions to the increasingly diverse and personalized learning demands of students. This systematic review presents a comprehensive synthesis of the integration status and practical applications of LLMs within programming education by analyzing 40 representative studies published between 2019 and September 2025. The overarching analytical framework is structured around three core research dimensions: typical LLM interaction patterns (RQ1), targeted learner adaptation strategies (RQ2), and implementation challenges alongside pedagogical risks (RQ3). In particular, existing applications are unified into four quintessential interaction roles: teaching assistant, feedback partner, task generator, and integrity monitor. To accommodate differentiated instructional needs, a 2-D learner adaptation framework is established, aligning teacher-facing and student-facing service paradigms with three distinct student cohorts: nonengineering majors (experience-oriented), non-CS engineering majors (application-oriented), and computer science majors (competence-oriented). Furthermore, this review critically discusses institutional challenges and instructor cognitive burdens while uncovering long-term risks of LLM-based applications in programming education. Ultimately, this work provides an actionable roadmap for curriculum designers and educators to transition from passive tool adopters to strategic architects of human–artificial intelligence (AI) collaborative learning environments.
Contribution: This article presents the development of a system-oriented electric vehicle (EV) powertrain course module that differs from traditional standalone control unit courses. It emphasizes real-time interaction and practical integration among subsystems, offering a novel framework for EV engineering education and serving as a foundational basis for future EV-related specialized courses, research, or advanced studies. Background: The powertrain and electrical architecture of EVs differ fundamentally from those of conventional fuel-powered vehicles. Therefore, during this industrial transformation, schools urgently need a prerequisite course on EV powertrain systems. However, most institutions currently lack such a course. With both industry experience and a background as an adjunct lecturer, the author developed and refined this course over three years through iterative curriculum development. Intended Outcomes: The course module is designed to provide students with a practical and system-level framework for EV powertrain systems. The teaching objectives are to enable students to understand the architecture and functionality of key EV subsystems and their interactions through in-vehicle network communication. Through the learning process, students will develop core competencies in embedded control, system integration, and teamwork, aligning with current industry practices in the EV sector. Application Design: This study adopts an outcome-based education (OBE)-guided, project-based instructional approach supported by systems-oriented and model-based learning, integrating theoretical instruction, modeling, simulation, and hands-on experimentation on a custom EV powertrain platform, with iterative refinement through formative assessment and statistical evaluation across three cohorts. Findings: Results show significant improvements in student learning outcomes across three cohorts, supported by descriptive statistics, one-way analysis of variance (ANOVA) with moderate to large effect sizes, enhanced performance in summative assessments, and consistent qualitative feedback indicating improved system-level understanding and engagement.
Contribution: This study presents the design and implementation of a didactic sequence that incorporates problem-based learning (PBL) to teach linear algebra in engineering. Its distinctive contribution lies in using a synchronization problem—typically addressed in advanced courses—as an authentic engineering context and the problem generator for the sequence. Background: Linear algebra is foundational in engineering education but is often perceived as abstract and disconnected from professional practice, which can negatively affect student motivation and learning. Instructional approaches that connect core concepts to relevant engineering contexts are therefore needed. Intended Outcomes: The approach aims to support conceptual understanding of linear algebra, enhance student motivation, promote theory–practice connections, and foster meaningful appropriation of information and communication technology (ICT). Application Design: The intervention integrates principles from the theory of didactical situations (TDS), collaborative learning, and purposeful ICT use within a didactic sequence developed using the analysis, design, development, implementation, and evaluation (ADDIE) instructional design model. A mixed-methods descriptive design was used. Findings: Data from a final project rubric and a postintervention questionnaire comprising Likert-scale self-report items and open-ended questions were analyzed across didactic, disciplinary, ICT integration, and technological appropriation dimensions. The findings provide evidence of successful student performance in the didactic sequence and positive perceptions of the relevance of linear algebra to engineering practice.
Contribution : This study formulates literacy tracing (LT), a joint sequential task that predicts response correctness and infers four-level evidence across six curriculum-defined mathematical literacies. It introduces LiteracyDataset and MoELT, a mixture-of-experts framework for LT. Background : Conventional knowledge tracing focuses primarily on binary correctness and knowledge-component mastery, providing limited support for multidimensional, curriculum-oriented assessment. Research questions (RQs) : How does MoELT perform compared with existing models? How do its key components contribute to performance? How sensitive is MoELT to task-weighting and hyperparameter settings? How accurately does it predict proxy-literacy levels, and how are the inferred profiles associated with subsequent same-literacy performance? Methodology : LiteracyDataset contains five months of mathematics assessment records from 22 112 students in 126 Chinese high schools. Exercises were annotated by three experts according to six mathematical literacies and three curriculum-defined requirement levels. MoELT integrates heterogeneous graph representation learning, graph contrastive learning (GCL), task-specific sequence updating, expert routing, and uncertainty-weighted multitask prediction. It was evaluated against nine knowledge tracing baselines across four grade-and-track subsets. Findings : MoELT achieved the highest next-response AUC and accuracy across all four subsets. Proxy-literacy prediction attained accuracies of 0.7070–0.7693 and quadratic weighted kappa (QWK) values of 0.3872–0.4977. Ablation and behavioral analyses supported the complementary graph design and showed positive associations between inferred profiles and subsequent same-literacy performance. These profiles represent interaction-level evidence rather than stable literacy attainment.
