Course placement in higher education is pivotal for student progression. Traditional advising can be effectively augmented by Large Language Models (LLMs) to ensure scalable guidance. Crucially, we propose a novel on-premises infrastructure that runs LLMs locally, ensuring strict student data privacy. We benchmarked four open-source LLMs against gold-standard responses from human advisors (n = 3) using 35 Information Systems student profiles, identifying GPT-OSS-120B as the optimal model. To guarantee real-world viability, we profiled the infrastructure's resource utilization across 112 sequentially processed cases. Additionally, we evaluated complex scenarios: undergraduate course substitutions (n = 10) and graduate prerequisite assessments (n = 10). Results indicate strong performance in substitutions but mixed outcomes for prerequisite assessments. We also determined the optimal model parameters for recommendation accuracy. To assess broader institutional utility, we expanded testing to include four additional disciplines: Marketing, Economics, Accounting, and Finance (n = 20 per major), demonstrating the system's potential for cross-disciplinary application. This study demonstrates that securely deployed, locally hosted LLMs can provide a privacy-compliant framework to assist academic advisors, though current performance highlights the necessity of continuous human oversight.
This study employs Systematic Literature and Network Analysis (SLNA) to explore game-based learning (GBL) in English language acquisition. A total of 1,774 publications on GBL and English as a Foreign Language (EFL), published between 2020 and 2025, were identified in Scopus, of which 444 met the eligibility criteria and were included in the final analysis. Citation network mapping and thematic clustering were conducted using VOSviewer to identify research trends and methodological developments. The findings reveal a growing emphasis on mobile technologies, artificial intelligence, immersive learning environments, and virtual reality applications in language learning, learner motivation, and vocabulary acquisition. Despite these advances, several limitations remain, including the lack of longitudinal studies examining long-term impacts, insufficient teacher training, and limited pedagogical integration of GBL practices. This study contributes to the existing literature by identifying emerging trends, gaps, and future research directions in the development and application of GBL in EFL contexts.
The effectiveness of AI tools in EFL learning depends not only on technology acceptance but also on learners’ self-regulated learning (SRL). However, prior research often treats learners as homogeneous and overlooks distinct user profiles. This study integrates the Technology Acceptance Model (TAM) and SRL to profile the use of AI tools (e.g., ChatGPT, Grammarly, and translation applications) by 365 Indonesian undergraduate EFL learners for independent learning. Using a descriptive-correlational design with a person-centered approach, data from a closed-ended Likert-scale questionnaire were analyzed through descriptive statistics, Pearson correlation, and two-stage cluster analysis (Ward's and K-Means). The results showed high TAM and SRL levels with significant correlations. Cluster analysis revealed three distinctive profiles: Highly Engaged (28.77%), Moderate (44.11%), and Low Engagement (27.12%). ANOVA confirmed significant differences (p < 0.001). Findings are context-bound and should not be generalized beyond the study setting, particularly across different educational and cultural contexts.
This study explores dialogic argumentation in secondary-level mathematics lessons and the potential of artificial intelligence (AI) and large language models (LLMs) to analyze these interactions. Its primary aim is to explore how and to what extent AI can assist in automating the complex process of analysing classroom interaction. Although argumentation and dialogic interactions share similarities, they differ in important ways: argumentation typically centers on justifying a single viewpoint, whereas dialogic interactions involve multiple perspectives, and not all dialogic interactions require justification. The study uses transcription of a secondary school mathematics lessons, employing human-validated coding protocols to create prompts for LLMs. Findings indicate that LLMs can reliably identify justifying moves but struggle to detect dialogic moves, often requiring repeated prompt refinements. The main challenge lies in clarifying the essential features of dialogic interactions for AI, highlighting the limitations and potential of automated analysis for educational research and practice.
Artificial intelligence (AI) technologies, including intelligent robots and adaptive learning systems, are increasingly integrated into education. However, educators often lack pedagogical guidance on their effective use in higher education. This study explores students’ self-regulatory behaviours and academic help-seeking practices while using ChatGPT during a web-programming task. An exploratory qualitative case study was conducted with 10 s-year computer programming students. Data from open-ended questionnaires and ChatGPT interaction records were analysed using content analysis. Findings indicated self-regulatory behaviours associated with foresight, performance, and self-reflection. Students mainly sought support for web-programming syntax, solution evaluation, and feedback. They used different help-seeking approaches and formulated prompts with varying detail and contextual information. Although several students expressed concerns about the accuracy of ChatGPT's responses, explicit verification was limited. The findings highlight ChatGPT's potential as a source of academic help during web-programming tasks while emphasising the importance of critically evaluating and adapting AI-generated responses.
