Large language models in education: How have research hotspots evolved? A longitudinal scientometric analysis

Objective. This study examined the evolution of China-affiliated scholarship on large language models (LLMs) in education following the ChatGPT debate. It analyzed the transformation of this scholarship into diverse research domains, including pedagogical, technical, and evaluative aspects. Design/Methodology/Approach. To this end, a comprehensive set of records was retrieved for the period spanning from 2018 to 2026, employing a meticulous approach that entailed the utilization of criteria such as LLM, education, Chinese institutional affiliation, and date. The screening cascade, as reported, reduced a total of 2,212 records to 539 articles, conference papers, and reviews that exhibited explicit LLM/model signals and education-centered titles. The analysis incorporated a multifaceted array of methods, including descriptive scientometrics, citation indicators, institutional productivity and affiliation summaries, controlled thematic coding, and seven-topic term frequency–inverse document frequency/non-negative matrix factorization (NMF) modeling. Results/Discussion. The final corpus, which was scheduled to commence in 2023, encompassed a total of 20 publications in 2023, 90 publications in 2024, 244 publications in 2025, and 185 publications in the partial-year 2026. Research attention shifted from broad discussions of opportunity, ethics, and adoption to more specialized topics such as teaching design, agentic/retrieval-augmented generation learning systems, empirical feedback and cognitive outcomes, and model-performance evaluation. Tsinghua University and East China Normal University were identified as the most prolific affiliations, with a total of 519 Chinese institutions represented in the corpus. The field exhibited rapid thematic differentiation, rather than merely volume growth. Conclusions. The publication record shows a broader range of research topics after 2023, particularly in instructional design, learning systems, feedback, and model evaluation. These findings reflect shifts in scholarly attention, not changes in classroom practice. Originality/Value. This study employs a multifaceted approach that integrates longitudinal NMF topic modeling with controlled thematic coding, aiming to elucidate the dynamics of scholarly attention. Sensitivity tests were conducted to assess the membership of the five leading controlled themes. The set's integrity remained uncompromised, irrespective of the exclusion of records from the partial-year 2026 period or the implementation of a more rigorous education-title criterion. The restriction to articles and reviews yielded four of the five themes. These tests did not assess the stability of NMF topics or institutional findings. The 2026 results describe a partial year through September 16.

Authors

Publication Details

Journal
Iberoamerican Journal of Science Measurement and Communication
Published
2026-11-21
DOI
https://doi.org/10.47909/ijsmc.406
Primary Topic
Artificial Intelligence in Education
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Large language models in education: How have research hotspots evolved? A longitudinal scientometric analysis

Canyi Xu, Lihong Ding, Ying Liu
Iberoamerican Journal of Science Measurement and Communication
Artificial Intelligence in Education
article

Large language models in education: How have research hotspots evolved? A longitudinal scientometric analysis

Canyi Xu, Lihong Ding, Ying Liu
article en

Abstract

Objective. This study examined the evolution of China-affiliated scholarship on large language models (LLMs) in education following the ChatGPT debate. It analyzed the transformation of this scholarship into diverse research domains, including pedagogical, technical, and evaluative aspects. Design/Methodology/Approach. To this end, a comprehensive set of records was retrieved for the period spanning from 2018 to 2026, employing a meticulous approach that entailed the utilization of criteria such as LLM, education, Chinese institutional affiliation, and date. The screening cascade, as reported, reduced a total of 2,212 records to 539 articles, conference papers, and reviews that exhibited explicit LLM/model signals and education-centered titles. The analysis incorporated a multifaceted array of methods, including descriptive scientometrics, citation indicators, institutional productivity and affiliation summaries, controlled thematic coding, and seven-topic term frequency–inverse document frequency/non-negative matrix factorization (NMF) modeling. Results/Discussion. The final corpus, which was scheduled to commence in 2023, encompassed a total of 20 publications in 2023, 90 publications in 2024, 244 publications in 2025, and 185 publications in the partial-year 2026. Research attention shifted from broad discussions of opportunity, ethics, and adoption to more specialized topics such as teaching design, agentic/retrieval-augmented generation learning systems, empirical feedback and cognitive outcomes, and model-performance evaluation. Tsinghua University and East China Normal University were identified as the most prolific affiliations, with a total of 519 Chinese institutions represented in the corpus. The field exhibited rapid thematic differentiation, rather than merely volume growth. Conclusions. The publication record shows a broader range of research topics after 2023, particularly in instructional design, learning systems, feedback, and model evaluation. These findings reflect shifts in scholarly attention, not changes in classroom practice. Originality/Value. This study employs a multifaceted approach that integrates longitudinal NMF topic modeling with controlled thematic coding, aiming to elucidate the dynamics of scholarly attention. Sensitivity tests were conducted to assess the membership of the five leading controlled themes. The set's integrity remained uncompromised, irrespective of the exclusion of records from the partial-year 2026 period or the implementation of a more rigorous education-title criterion. The restriction to articles and reviews yielded four of the five themes. These tests did not assess the stability of NMF topics or institutional findings. The 2026 results describe a partial year through September 16.

Iberoamerican Journal of Science Measurement and CommunicationVol. 6
Artificial Intelligence in Education
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon