Javascript is required
Search
Volume 3, Issue 4, 2025

Abstract

Full Text|PDF|XML
To address the limitations of traditional policy instrument analysis—such as labor-intensive coding, high subjectivity, and time-consuming procedures—this study develops a policy instrument analysis framework that integrates large language models (LLMs) and proposes a LLM-driven analytical workflow comprising six stages: case repository construction, policy instrument selection, content element generation, clause-level coding, reliability and validity testing, and quantitative analysis. Using governance texts on teachers’ ethical misconduct from 27 universities specializing in finance and economics as the empirical context, the study employed DeepSeek-R1 to identify policy instruments, classify content elements, perform clause-level coding, and conduct two-dimensional cross-tabulation analysis. The results indicate that these governance texts exhibit pronounced regulatory, procedural, and accountability-oriented characteristics, while also revealing a structural imbalance marked by strong front-end norm construction and relatively weak back-end remedial mechanisms. Overall, the proposed framework improves the efficiency and consistency of policy text analysis and provides a novel technical pathway for methodological innovation in education policy research.

Abstract

Full Text|PDF|XML

ChatGPT, a widely used generative AI tool, has incessantly attracted significant attention from researchers seeking to understand the factors that influence its adoption in higher education. This study examined the determinants of ChatGPT adoption among university students in Yogyakarta, Indonesia, one of the largest educational centers with more than 100 higher education institutions. Drawing on the value-based adoption model (VAM), the study incorporated three AI-related factors, i.e., AI self-efficacy (ASE), perceived academic value (PAV), and privacy concerns (PC) to appropriately explain students’ behavior in AI adoption. A hybrid analytical approach combining partial least squares structural equation modeling (PLS-SEM) and machine learning (ML) techniques, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN) with SHapley Additive exPlanations (SHAP)-based interpretation, was employed to test ten hypotheses with survey data collected from 484 students across five selected universities in Yogyakarta. The results indicated that perceived value (PV) (β = 0.453) and adoption intention (β = 0.521) were the strongest predictors of actual ChatGPT usage. Among the benefit-related factors, perceived usefulness (PU), perceived enjoyment (PE), and ASE significantly enhanced PV, whereas PC (β = −0.213) represented the most influential barrier to adoption. The ML models produced consistent findings, with XGBoost achieving the highest predictive performance (AUC = 0.912). SHAP analysis further highlighted PV and PC as the most significant variables. By extending VAM with AI-specific constructs and integrating SEM with ML techniques, this study contributes to an enhanced understanding of generative AI adoption in higher education and offers actionable insights for policymakers and university administrators in support of responsible AI integration in Indonesian universities.

Abstract

Full Text|PDF|XML
Mental health problems like eco-anxiety were caused by the rise of climate change and constituted a major concern within higher education settings. Eco-anxiety, a chronic fear of environmental catastrophe, profoundly impacts students’ psychological well-being and academic engagement. This qualitative case study interviewed five professors teaching English as a Foreign Language (EFL) at the University of Turku in Finland and reported how they perceived eco-anxiety and managed its integration into language education. Data collection consisted of semi-structured interviews, accompanied by eight-hour structured and non-participant classroom observations in five different classes over a full week. To ensure analytical rigor, a 30% independent cross-coding verification protocol was completed on the transcripts, and a matrix analysis was applied to compare stated educational beliefs with field observations. The findings revealed a marked theory-practice gap: while EFL professors demonstrated high conceptual awareness of eco-anxiety and recognized it as a valid student’s response, its active pedagogical integration remained significantly limited. Data from objective observation reported an almost total absence of explicit discourse on climate or eco-anxiety in daily teaching routines. Matrix triangulation substantiated that the goodwill of the individual educator was systematically hampered by severe institutional barriers, primarily curriculum overload, limited teaching time, and a lack of formalized institutional or financial support. When professors addressed isolated sustainability tasks through frameworks such as the United Nations Sustainable Development Goals (SDGs), students experienced an intense emotional trajectory that could temporarily heighten their affective filters and inhibit language production. However, when properly supported, these discussions fostered long-term vocabulary growth and communicative agency. These findings urge curriculum architects to design structured resources and institutional frameworks that seamlessly integrate emotional and environmental education into the core language curriculum.
- no more data -