Abstract Although recommender systems have the potential to advance self-regulated learning (SRL) in online environments, implementation challenges impede their development. To systematically investigate the barriers to the effective use of recommender systems to enhance SRL, this study developed a three-tiered analytical framework that considers challenges in three areas: data and experiments, user experience/interface design, and technical factors. A systematic search of five bibliographic databases identified 33 empirical studies. The findings revealed the following key challenges in the three analytical levels: (a) data and experiments: scarcity, validity, and arrangement; (b) user experience and engagement: poor matching and inadequate scaffolding for SRL (e.g., weak feedback mechanisms) and suboptimal interface and function designs; and (c) technical factors: lack of stability and security of recommender systems and limitations in modeling complex learning behaviors and balancing personalization with pedagogical goals. These challenges interact dynamically to form a self-reinforcing recursive constraint, where poor user experiences lead to disengagement, resulting in sparse, low-quality data that further constrains algorithmic capabilities. Current algorithmic approaches typically address isolated problems, failing to break this cycle by overlooking these systemic interdependencies. This underscores the critical need to move beyond isolated fixes and develop comprehensive, pedagogically-aware algorithmic architectures. To address this, this study proposes a four-phase development roadmap for creating an SRL-Scaffolding Architecture. This roadmap integrates multi-modal data validation, hybrid algorithmic intelligence combining emotional and cognitive modeling, adaptive scaffolding interfaces, and a continuous human-in-the-loop feedback cycle. By mapping these interconnected challenges and proposing a structured architectural framework, this study proposes roadmap to address these barriers and develop recommender systems that effectively empower SRL in online environments.

Abstract This longitudinal study, grounded in Dynamic Systems Theory (DST), explores how language teachers’ integration of AI tools evolves over an 18-week period, revealing AI adoption as a complex pedagogical transformation rather than a simple technological shift. Drawing on these findings, the research introduces two models: (1) the DST-informed Pedagogical Content Knowledge (PCK) model, which specifies AI-empowered PCK by detailing the five domains of knowledge teachers draw upon, and how these inform teaching practice and scaffolding strategies for personalised student learning; and (2) the AI-in-PCK stage framework, which maps the trajectory of AI adoption, illustrating how teachers’ concerns and practices evolve from initial exploration and experimentation to strategic integration and ongoing learning, while responding to classroom realities and student feedback. Together, these models illuminate adaptive, multifaceted changes in PCK and teaching practice, highlighting how AI integration shapes decision-making and professional growth. The findings underscore critical implications for designing flexible, context-responsive professional learning and systemic support strategies, particularly in under-resourced rural contexts, and provide a foundation for future AI-in-PCK research.

Abstract Students often struggle to stay engaged and effectively regulate their learning in online environments, which can negatively impact their learning experiences. Despite the established importance of self-regulation of learning skills (SRLs) in maintaining engagement, many students face significant challenges in developing and implementing these skills due to a lack of adequate feedback. This is primarily due to tutors' high workloads and the difficulties inherent in engaging students in online settings. This study examines the impact of Learning Analytics (LA)-driven interventions to improve students' SRLs in online learning environments. Specifically, it compares the impact of feedback from Generative Artificial Intelligence (GenAI) and human tutors in a nine-week statistics course delivered via MOODLE at a higher education institution, employing LA and clustering techniques to model SRLs based on Ye & Pennisi’s (2022) framework. In a quasi-experimental design, participants with varying SRLs were assigned to either a tutor-feedback or GenAI-feedback group. Feedback readability and reliability evaluations indicate that LA-driven GenAI-produced feedback was significantly more readable than human tutor feedback (p < 0.01) and demonstrated higher reliability than tutor-generated feedback. Results show that students in the low SRLs cluster receiving GenAI feedback exhibited statistically significant improvements in goal-setting skills (p < 0.05) and overall SRLs levels (p < 0.05) compared to the tutor-feedback group. In contrast, no significant differences were observed among high SRLs cluster students. This study underscores the potential of LA-driven GenAI feedback to be able to develop tailored, scalable feedback, improving SRLs performance of low SRLs students in online higher education contexts. Future research should explore these effects across diverse student groups and investigate the collaborative potential of semi-automated feedback systems that include tutors.

Abstract The present study investigated the interplay of cognitive cues and social cues in instructional videos on students’ knowledge acquisition and sense of social presence. 312 preservice teachers took part in the study. Out of these, 238 participants were randomly assigned to one of three experimental groups receiving an asynchronous online video lecture on three research methods topics, featuring either static slides with voiceover (minimal signaling), dynamic slides with voiceover (moderate signaling), or dynamic slides with additional instructor signaling (maximum signaling). 74 participants served as a control group and completed pre- and posttests without attending the course. Perceived difficulty, invested effort, motivation, and notetaking were included as covariates. Dynamic slides did not improve knowledge acquisition compared to static slides. However, instructor signaling enhanced knowledge acquisition when perceived difficulty was high. Furthermore, instructor signaling significantly increased perceptions of social presence for two of the three investigated topics, highlighting the importance of social cues.

Abstract This investigation considers the relationship between test scores and a sorting task conceptual macrostructure measure based on topic-level term-term distances as Pathfinder networks. In Study 1 (n = 255), grade 7 Chinese students completed a sorting task 1 month after the traditional in-class lessons and exam. In Study 2 (n = 220), grade 8 students completed the sorting task immediately after self-directed study of a history text. In addition, a month later 68 of the students in Study 2 were further instructed to write a short essay about this content. Study 1 results showed significant correlations between the sorting task macrostructure network measures and both lesson and unit test scores. Study 2 obtained the same significant correlations between sorting task macrostructure network measures and performance on tests. In addition, in Study 2, essay conceptual networks of historical content were better for the high prior knowledge students. Both the sorting task and the essay writing task measures can complement traditional exam measures so that conceptual knowledge structure aspects of students’ learning can be identified for formative and summative purposes.

Abstract Mobile learning can positively impact learning in different aspects, but the retention rate of mobile learning applications could be better. Based on the Technology Acceptance Model and the updated DeLone and McLean Information System Success Model, this study develops a novel model to examine the determinants of learners’ acceptance of mobile learning outside the classroom. Learning outside the classroom refers to voluntary learning activities that occur beyond the physical classroom and scheduled instructional time, including activities performed by both students and non-students (e.g., those not currently enrolled in educational institutions). Six hundred eighty-one adults in the U.S. participated in this study. We utilized structural equation modeling for data analysis. Results indicate that two quality dimensions, namely system quality (mobility and compatibility) and service quality, and two learners’ beliefs, namely perceived usefulness and perceived ease of use, play an essential role in m-learning acceptance outside the classroom.