Qi Xu

Learning engagement is an overarching multifaceted construct that includes behavioral, cognitive, and emotional dimensions. The static subjective measurements may lead to inaccurate results. Considering the authenticity and reliability of body signals, this study investigated the types of body cues used in measuring learning engagement and their relationship with engagement. By examining non‐verbal visual, speech, vital, and hybrid cues, we find that non‐verbal visual cues offer moderate‐to‐high accuracy at low cost and are widely applicable; vital signals achieve high accuracy but are intrusive, expensive, and largely lab‐confined; speech cues perform moderately in communication‐rich settings; and hybrid cues attain the highest potential accuracy through multimodal fusion, albeit with added complexity. We also find that measuring both the levels and states of engagement contributes to a better understanding of learning engagement. However, “levels” may require more subjective judgment, whereas “states” may be less straightforward to interpret. Moreover, collecting more contextual cues helps to measure learning engagement accurately. Because the engagement is multi‐sensory triggered, hybrid cues make the measurement more realistic and reliable due to more auxiliary cues. These findings offer practical guidance for selecting suitable measurement tools across educational contexts and highlight the necessity of multimodal context‐aware systems for precise and objective perception of learning engagement.

Generative AI (GenAI) is reshaping the way students acquire knowledge by providing personalized and interactive learning opportunities. However, despite its growing adoption, the mechanisms through which these interactions influence students' engagement and well‐being in GenAI‐assisted learning environments (GenAI‐ALEs) remain underexplored. This study surveyed 750 university students in central China to investigate how diverse interaction forms (student – teacher, student – content, student–student and student – AI) affect their engagement and well‐being in GenAI‐ALE, mediated by multi‐dimensions of self‐efficacy: academic, social and emotional components. Partial least squares structural equation modelling (PLS‐SEM) was employed to test the hypothesized relationships, while fuzzy‐set qualitative comparative analysis (fsQCA) was used to uncover complex causal configurations, which offered a richer perspective on the underlying mechanisms. Results revealed that the multi‐dimensional self‐efficacy mediated the effects of student–student and student – AI interactions on both engagement and well‐being; academic self‐efficacy mediated the link between student – teacher interaction and these outcomes and emotional self‐efficacy mediated the relationship between student – content interaction and the same outcomes. Furthermore, the fsQCA identified multiple causal pathways that influence student engagement and well‐being. The findings yield valuable theoretical and practical implications for designing effective GenAI‐ALEs to foster students' engagement and well‐being.

Cognitive engagement involves mental and physical involvement, with observable behaviors as indicators. Automatically measuring cognitive engagement can offer valuable insights for instructors. However, object occlusion, inter-class similarity, and intra-class variance make designing an effective detection method challenging. To deal with these problems, we propose the Object-Enhanced–You Only Look Once version 8 nano (OE-YOLOv8n) model. This model employs the YOLOv8n framework with an improved Inner Minimum Point Distance Intersection over Union (IMPDIoU) Loss to detect cognitive engagement. To evaluate the proposed methodology, we construct a real-world Students’ Cognitive Engagement (SCE) dataset. Extensive experiments on the self-built dataset show the superior performance of the proposed model, which improves the detection performance of the five distinct classes with a precision of 92.5%.