Task interruptions are pervasive in digital learning environments for science, technology, engineering and mathematics (STEM), yet the support required for effective interruption recovery among learners differing in self-regulation remains poorly understood. Here, we developed a three-layer recovery-support framework comprising Locate, Recall and Predict (LRP), yielding 27 combinations of support strategies. Using a mixed-design experiment, we studied 60 higher-education students from programs in or related to Industrial Engineering in an anonymized higher-education setting, all with prior experience in programming-related learning tasks. We combined linear mixed-effects modeling (LMM) with XGBoost–SHAP (SHapley Additive exPlanations) analysis to examine overall support effects and group-specific patterns of feature contribution. Support-related effects were most consistent for recovery time. Self-regulation group had a significant effect on recovery time (χ² = 421.69, p < .001), with mean recovery times of 18.6, 28.2, and 36.8 s in the high-, medium-, and low-SR groups, respectively. Further analyses showed that the relative contributions of recovery-support components differed across self-regulation groups. These findings indicate that recovery outcomes vary across learners and recovery-support components, with these patterns differing by self-regulation group. Digital learning platforms could therefore use learner profiles and observed recovery states to select among Locate, Recall, and Predict cues rather than applying the same recovery support uniformly to all learners.
As generative artificial intelligence (GenAI) becomes increasingly embedded in higher education, understanding how students interpret AI-assisted feedback is important for developing effective digital pedagogy. Rather than focusing solely on whether GenAI improves learning outcomes, this study examined how students’ perceptions of AI-assisted feedback were associated with perceived learning effectiveness through trust-related and motivational processes. Three independent but conceptually linked studies combined mixed-design analysis of variance, structural equation modeling (SEM), and platform-experience group comparisons. Study 1 compared pre–post changes in learning-related perceptions between GenAI-assisted and instructor-feedback conditions. Study 2 evaluated a dual-pathway model linking user perception of AI-assisted feedback (UPAF) with perceived learning effectiveness (PLE). The results showed two significant indirect association patterns: UPAF was associated with PLE through trust in AI-generated content (TAC) and engagement propensity (EP), and through perceived teacher facilitation role (PTFR) and pedagogical trust (PT). By contrast, the direct structural associations from UPAF to PT, EP, and PLE were not statistically significant. Study 3 further showed that students with greater prior platform experience reported higher levels of UPAF, TAC, and PTFR, indicating experience-based differences in the interpretation of trust-related and pedagogical cues. Overall, the findings support an association pattern consistent with the proposed dual-pathway framework and offer practical implications for designing more transparent, trust-sensitive, and pedagogically interpretable feedback systems in higher education.
Machine translation (MT) has become integral to foreign language learning, yet users’ motivations, behaviors, and ethical engagement remain underexplored. This mixed-method study examines how Chinese university students adopt and perceive MT. Drawing on exploratory interviews (N = 48), a questionnaire was developed and administered to 1, 004 students. Latent class analysis revealed a monotonic hierarchy of three distinct user types: Two-purpose users (5.6
The rapid emergence of generative artificial intelligence (GenAI) tools challenges traditional assessment methods in higher education, particularly within STEM disciplines. This study explores (1) how integrating GenAI into student exams affects performance, (2) to what degree students critically verify GenAI-generated responses, and (3) how students use GenAI during exams. Employing a within-subjects experimental design, 95 undergraduate STEM students from two engineering and computer science courses completed two comparable short in-class exams: one with and one without GenAI assistance. Data were collected via exam scores to measure performance and a post-exam questionnaire to analyze usage patterns and verification behaviors. The results showed significant overall performance improvement, with larger performance gains observed among students with lower baseline performance, whereas students with higher baseline performance showed minimal improvement. Furthermore, after accounting for baseline performance, students who reported full verification showed significantly greater improvement in exam performance than those who reported partial verification. Usage patterns indicated that GenAI was primarily used for task-oriented solution-seeking and for verifying accuracy, rather than for deepening conceptual understanding. Theoretically, the observed performance pattern may be interpreted through principles of cognitive load theory and the expertise reversal effect. Practically, it calls for a paradigm shift in assessment, moving from evaluating knowledge recall to evaluating critical verification competence and evaluative judgment in the era of intelligent technologies.
Learning Analytics dashboards (LADs) are widely promoted for enhancing instructional decision-making, yet they continue to fall short of teachers’ needs and classroom realities. Despite calls for human-centred design to address such necessity, evidence on sustained teacher involvement across the design process is limited, and little is understood about how teachers conceptualise pedagogically meaningful data or negotiate control in analytics-supported instruction. This study investigates how teachers can be engaged as pedagogical co-designers in the development of an LAD designed as a “hub LAD” integrating student data from multiple educational systems for Maths in primary education. We conducted a multi-phase qualitative phenomenological study involving six Dutch primary school Math teachers. The multi-phase process involved interviews, design-thinking workshops and sketching activities for brainstorming, conceptualisation and initial evaluation of a low-fidelity prototype. Findings reveal challenges related to fragmented data and limited visibility into student learning, reflecting a misalignment between what LADs present and the pedagogical, social, and contextual knowledge teachers draw upon. Teachers articulated a demand for action-oriented analytics beyond performance indicators to include socio-emotional, behavioural and analogue dimensions of learning, challenging assumptions about what counts as legitimate pedagogical data. Three core design requirements emerged: (1) monitoring and alerting, (2) pedagogically grounded student clustering, and (3) advisory decision-support that preserve teacher autonomy. These revealed relational stances toward automation, forming a typology of teacher-LAD relationship in which LAD automation and teacher control is situationally negotiated. The current study advances understanding of how sustained co-design can reshape LADs toward pedagogically meaningful, context-sensitive, and action-oriented educational solutions.
