Mining personalized learning feature labels from student reflection reports: A computational approach to adaptive education

This research introduces an innovative computational framework that automatically extracts multidimensional personalized feature labels from student course reflection reports. Unlike traditional learner profiling methods that rely on questionnaires and standardized assessments, our approach leverages natural language processing, sentiment analysis, and unsupervised learning to mine naturally occurring student-generated text. The key innovation lies in our integration of computational linguistics with educational theory to identify five distinct learner profiles (Academic Analytical, Innovative Thinker, Collaborative, Optimistic Active, and Expressive Rich Learner) along with specific trait labels that capture unique cognitive, emotional, and social dimensions of learning. We demonstrate the framework’s effectiveness through experimental validation with both student self-assessment (78.2% agreement) and instructor evaluation, confirming that algorithmically-determined classifications strongly align with human perceptions. Our methodology establishes formal mapping relationships between extracted personalized labels and specific instructional strategies, enabling educators to tailor teaching approaches to individual learning profiles. By analyzing unstructured text that students naturally produce during their educational journey, we achieve non-intrusive learner profiling that provides insight into student learning characteristics. The methodology we’ve developed establishes a foundation that could potentially be extended to capture learning development over time in future longitudinal studies with larger samples. This work contributes to the advancement of adaptive education by bridging the gap between computational data analysis and practical pedagogical interventions, ultimately supporting the development of more responsive and personalized learning environments.