Knowledge Evolution and Paradigm-Level Change in AI-in-Education Research: A Scientometric Mapping and Critical Review (2015–2025)

Abstract

Bibliometric commentary increasingly frames the rise of generative artificial intelligence (AI) as a paradigm shift in AI-in-education (AIED) research, yet rarely distinguishes publication growth from genuine reconfiguration of the knowledge base or of teacher–learner roles. This study maps knowledge evolution in AIED from 2015 to 2025 by integrating four analytical strands across Web of Science and Scopus: scientometric mapping (including keyword and reference bursts and co-citation analysis), structural topic modeling (STM), design-weighted coding of full texts, and triangulation across these strands. Two transition windows emerged. The weaker 2019–2020 window (Jensen–Shannon divergence, JSD = 0.058; retained under the permutation rule but not under a stricter interval-based rule) reflects a research-front transition toward learning analytics and prediction, accompanied by knowledge-base reconfiguration but no detectable change in teacher–learner roles. The 2022–2023 window (JSD = 0.187, the period maximum) coincides with the generative-AI surge and couples a newly formed co-citation community with a topic-specific rise in coagency whose direction is clear but whose magnitude is imprecisely estimated. Because these effects concentrate in 36.6% of the weighted corpus and recent citations remain right-censored, we classify the change, under our pre-specified labeling scheme, as localized paradigm-level reconfiguration rather than a field-level paradigm shift. The label is an operational classification within that scheme, not an independent determination that a paradigm shift has occurred.