Answers are grounded in this paper and cite their sources.
Smart learning environments increasingly rely on data-driven methods to support individualized instruction. This paper proposes a multidimensional learner persona modeling framework that integrates behavioral, cognitive, and motivation-related indicators derived from LMS logs and self-report proxies. Using the Canvas open dataset, we construct five learner indicators (active days, number of interactions, course completion rate, user type, and learning expectations) and evaluate clustering performance under different distance metrics. We implement K-Means++ with Euclidean distance and a covariance-aware variant using Mahalanobis distance computed from the standardized feature covariance matrix, and compare them with density-based (DBSCAN) and probabilistic baselines. Results show that Mahalanobis K-Means++ yields improved clustering quality (Silhouette = 0.4277; Davies–Bouldin = 0.8118; Calinski–Harabasz = 1906.56) compared with Euclidean K-Means++ (Silhouette = 0.3281; Davies–Bouldin = 0.8279; Calinski–Harabasz = 1472.13). Four interpretable personas are identified and mapped to differentiated instructional interventions across task, activity, resource, and evaluation dimensions, with feasibility assessed via expert review. The study highlights the practical value of distance-metric selection for learner segmentation and provides a reproducible pipeline for persona-driven instructional support.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: