Temporal Semantic Flow Networks for Analyzing Topic Evolution in Educational Data Mining

This study aims to provide a comprehensive longitudinal analysis of the evolution of Educational Data Mining (EDM) research from 2014 to 2024, with a particular emphasis on learning analytics and student success prediction. We have developed and implemented the Temporal Semantic Flow Networks with Attention-based Topic Evolution (TSFN-ATE) methodology, which reconceptualizes the evolution of scientific discourse as continuous semantic flows. This framework integrates multi-scale temporal attention mechanisms and dynamic network representations to analyze Scopus-indexed keyword data from 436 EDM articles. The analysis identified significant transformations within the discourse of EDM, with a 333% increase in research intensity related to predictive modeling and an unprecedented 2,650% growth in the application of deep learning from 2014 to 2024. Deep learning attained the highest semantic flow score (0.94), signifying its successful integration across various EDM subfields, while machine learning emerged as the central methodological bridge, exhibiting the strongest network centrality (0.74). Multi-scale temporal attention analysis revealed differential patterns across time scales, with machine learning receiving the highest recent attention (0.85 at a 1-year scale), whereas learning analytics demonstrated sustained long-term influence (0.75 at a 5-year scale). Concept drift detection identified four major paradigm shifts, with the 2018-2019 deep learning integration representing the most significant conceptual disruption (semantic stability index 0.45). Collectively, these four identified paradigm shifts — from traditional machine learning through deep learning integration to the emergent ethical turn — reveal that EDM is not simply accumulating methods but undergoing continuous conceptual reorganisation; understanding and anticipating these shifts is therefore essential for researchers, institutions, and policymakers seeking to align educational technology development with evolving scientific and societal priorities.