Problem-based learning (PBL) enables students to solve real-world problems collaboratively. However, the effectiveness of PBL depends heavily on problem identification and appropriate group formation, as both can negatively affect learning outcomes. Recent research has explored the use of learning analytics-driven algorithms to support optimal group formation. In this paper, we propose a pairwise learner modelling approach to examine how individual learner behaviors relate to performance within PBL groups in an undergraduate distributed computing course. Multimodal learner interaction data, including eye gaze, clickstream interactions, and text-based responses, were collected through an online multimodal analytics enabled learning platform. These data were analyzed to compute pairwise similarity indices among group members and to examine their relationship with the quality of group solution design and implementation. The findings aim to provide insights into how behavioral similarity among group members is related to learning outcomes in PBL settings.
Recent advances in AI-based image generation offer new possibilities for visual feedback that may encourage content elaboration and semantic reflection. This study compares the effects of synthesized image feedback and text-based automated writing evaluation (AWE) feedback on English as a foreign language learners’ revision behaviours, linguistic accuracy, and changes in text length and syntactic feature counts. Sixty-seven Japanese undergraduate students completed picture description tasks using a web-based writing system that provided either synthesized image or AWE feedback. Analyses included 64 writing samples containing valid revisions from 46 participants. Revision coding, rubric-based assessment, and mixed-effects models were used to examine changes between initial and final submissions. Content-level revision occurred at similar rates in both conditions, whereas surface-level correction was descriptively more frequent under AWE feedback. AWE feedback produced significantly greater gains in linguistic accuracy. However, between-condition differences in content-score gains and changes in text length and syntactic feature counts were not statistically significant. Qualitative examples suggested that synthesized images could prompt descriptive elaboration and help learners identify mismatches between intended and expressed meaning without explicit vocabulary suggestions. These findings suggest that synthesized image feedback can support meaning-focused revision, while AWE provides stronger support for linguistic accuracy. Combining the modalities across drafting and editing stages warrants further investigation, as their combined effects and implications for long-term writing development remain untested.
This study compares two integrated concept mapping designs within a collaborative problem solving (CPS) setting. The first employs a component-provided reconstruction-based (CPRB) concept mapping approach with real-time automated diagnostic feedback (Kit-Build), while the second uses a component-provided flexible-linking (CPFL) approach with delayed human-mediated feedback (Figma). In both conditions, learners received the same concept-node and linking-word components derived from a lecturer-defined goal map, but the two conditions differed in how learners were expected to use these components. In addition, the conditions differed in mapping structure, feedback timing, and feedback type. Rather than isolating a single factor, the study evaluates the combined effects of these bundled instructional designs relative to a flexible-linking baseline commonly used in an Indonesian university, where feedback is typically provided one week after task completion. Participants were 78 students in the experimental group (EG) and 40 in the control group (CG), with identical content across one preliminary individual session and three paired case-based sessions. Using an explanatory sequential mixed-methods design, data included concept map scores, pre/post tests, summative assessments, and qualitative reports. Results showed that the CPRB configuration was associated with higher concept map scores across sessions than the CPFL configuration, with a large effect size in three sessions and a medium effect size in one. Post-tests showed medium to large effects, and summative results favored the CPRB condition (r = 0.8064). Differences in students’ CPS-related perspectives between the two conditions were minimal, with broadly similar patterns across groups. Overall, both CPRB and CPFL concept mapping were perceived as similarly acceptable, while the CPRB design showed comparatively higher learning performance across sessions. This indicates that CPRB may have potential to better support learning outcomes, although further research is needed to confirm this pattern and to better understand the mechanisms underlying these differences.
Student-facing learning analytics dashboards support formative assessment, yet many present simple metrics that learners struggle to interpret and act on. We propose Social Process Awareness (SPA): using peers’ learning processes as social feedback to reflect on one’s own process. We designed an SPA-support dashboard for long-break assignment and conducted a two-step study in a Japanese junior high. From learning logs of 108 seventh graders, time-series clustering of progress identified Low Engagers, Early Finishers, and Consistent Finishers, and clustering of effort (reattempt rate and final accuracy) identified Proficient, Struggling, and Persistent students. We then deployed the dashboard with 164 seventh and eighth graders and analyzed system logs, reattempt performance, and exam outcomes by achievement goal orientation. Dashboard use did not predict exam scores, but was positively associated with incorrect-to-correct improvement during reattempts, strongest for very high goal orientation. Medium/high users (≥ 6 uses) improved more than no/low users. These findings suggest SPA as a promising process-focused social dashboard approach and motivate tailoring feedback to learner characteristics.
This study examines the challenges digital educators and leaders across Europe face in driving systemic change amid rapidly evolving technologies, and how these challenges relate to the digital divide. An exploratory sequential mixed-methods design combined semi-structured interviews with educators and leaders in Ireland, France, Spain and Latvia with a subsequent survey built from the interview themes; findings are reported thematically across the four systems rather than country by country. Despite increased technology use, participants described persistent inequities in infrastructure, digital skills and curriculum integration, and a marked gap between the importance they attached to key leadership competencies and their confidence in exercising them, particularly around procurement, legislation, financial planning and strategy. Leaders also identified obstacles that current digital-divide theory does not capture: the demands of planning and funding, resistance to change, and the accelerating pace of innovation. While digital tools were seen as supporting inclusion, concerns about wellbeing, distraction and the erosion of core skills also emerged. The study argues that technocentric approaches to modernisation deepen rather than close these divides, and proposes a fifth level, innovation capacity: the institutional ability to anticipate, evaluate and integrate technologies whose form is not yet stable. Strengthened digital educational leadership is presented as the mechanism through which more equitable, future-focused systems can be built.
