A novel fuzzy clustering approach with transition matrix for explainable evaluation of social media-based digital literacy interventions

Abstract

Assessing the effectiveness of digital literacy interventions often relies on raw score comparisons or hard classifications, which may obscure nuanced changes in conceptual understanding and provide limited interpretability. Traditional approaches fail to capture the probabilistic and fuzzy nature of learning progression and do not support transparent analysis of how learners transition across conceptual clusters over time. This study proposes an explainable evaluation framework that integrates fuzzy clustering with a fuzzy transition matrix to model the redistribution of aggregated membership values between pretest and posttest conceptual clusters. The framework applies Fuzzy C-Means (FCM) to derive soft cluster memberships and constructs a transition matrix that represents probabilistic learning progression in a linguistically interpretable form. Unlike conventional methods, this approach enables the analysis of gradual transitions across levels of proficiency rather than binary outcomes. The model was applied to real-world educational data from control and experimental classes, the latter of which received a social media-based instructional intervention. Results indicate that the control class exhibited downward or stagnant patterns, particularly among high-performing learners, while the experimental class showed more coherent upward cluster transitions among low- and moderate-level learners. By enabling interpretable modeling of pre–post cluster transition patterns, the proposed framework contributes to the advancement of explainable machine learning in education. It also highlights the potential of social computing platforms to foster scalable, data-driven digital literacy development.