Source-linked AI summary
How Does Science Education Research Respond to Sociopolitical Change? A BERTopic Analysis of Korean Research
Jibeom Seo, Junghyo Jo, Sonya N. Martin, Taejin Byun
TL;DR
Prior science education research has emphasized topics and trends more than the external conditions shaping research. This study analyzes Korean science education publications with BERTopic and finds that some policy- and practice-anchored topics track sociopolitical change while more discipline-grounded topics remain comparatively stable.
Problem
The external conditions shaping science education research agendas remain less examined than topics and trends themselves.
Method
The study analyzes topic proportions and temporal trends using BERTopic, supplemented by Mann–Whitney U tests and Cliff’s δ effect sizes.
Results
Policy- and practice-anchored topics track external change more closely, whereas topics grounded in established academic agendas show relatively constant publication trends.
Takeaways & Limitations
The findings support testing whether the same topic is anchored differently across national contexts rather than comparing only its frequency.
Takeaways & Limitations
About one third of papers were excluded as outliers, and external data cannot provide direct evidence of causal relationships.
Abstract
from arXiv · showhide
Research fields do not evolve in isolation: their questions and priorities shift with policy, curriculum reform, and broader social change. Analyzing published literature can reveal not only how a field matures but also how it responds to these conditions. Prior work in science education has focused on identifying research topics and their trends, but paid less attention to the external conditions in which research is produced. We examine Korean science education research from 2008 to 2025, a case in which centralized curriculum revision, government education initiatives, and demographic decline are prominent. Using BERTopic, an embedding-based topic modeling technique, we identify major topics and temporal trends, and analyze their associations with selected sociopolitical factors. We interpret each topic and distinguish three groups: sociopolitical, subject-specific, and student-related topics. Within the first group, science teacher professionalism and curriculum implementation, science education for gifted students, and STEAM education show the strongest associations with sociopolitical conditions, such as government policy initiatives and declining enrollment in science-gifted education, whereas digital-based science education does not. The subject-specific and student-related groups, by contrast, show no comparable movement and are not linked to the external indicators we examine; this pattern is interpreted as reflecting stronger disciplinary grounding. Taken together, these patterns suggest that a topic's anchoring to policy and practice or to academic disciplines shapes how closely it tracks external change. This helps explain why some research agendas move with their national context while others hold steady, and why the same topic may develop differently across countries.
Overview of Korean science education · Historical Background and Curriculum System · Educational Practices and Challenges
Korean science education is shaped by a centralized, regularly revised national curriculum and a strong emphasis on science achievement. However, students show comparatively weak affective attitudes toward science amid test-oriented practices, prompting the government to introduce STEAM education in 2011.
- Overview of Korean science education: Korean science education is presented as the context for examining how the field responds to national policy and broader educational conditions.This overview establishes the background for the study.
- Historical Background and Curriculum System: After the Korean War, training a skilled STEM workforce became a national priority tied to reconstruction and development.This priority shaped subsequent science education policy.
- Historical Background and Curriculum System: The centralized national curriculum has allowed national policy directions to exert decisive influence on Korean science education.Korea issued its first curriculum in 1954, and the 2022 Revised National Curriculum was announced as the 11th version.
- Historical Background and Curriculum System: Curriculum revisions adjust subjects and content to emerging needs, but school implementation typically lags announcement by a few years.This creates a temporal gap between policy revision and classroom practice.
- Historical Background and Curriculum System: School science is integrated in Grades 1–2, taught as a mandatory core subject in Grades 3–9, and specialized through credit-based electives in Grades 10–12.Grades 11–12 offer general and career electives across physics, chemistry, biology, and earth science.
- Educational Practices and Challenges: Korean students rank among the world’s top performers in science achievement but report low confidence and little enjoyment in science learning.Their strong problem-solving skills coexist with comparatively low affective attitudes toward science.
- Educational Practices and Challenges: Knowledge-based assessment and intense college-admission competition encourage test-taking strategies rather than authentic inquiry.These prevailing practices are identified as contributing to students’ weak affective attitudes toward science.
- Educational Practices and Challenges: 2011 marked the introduction of government STEAM education to raise affective attitudes toward science and encourage STEM career choice.The initiative integrates humanities and arts into STEM to enhance student interest and understanding in science and technology.
Overview of Methods in Science Education Literature Research … Conceptual diagram of the BERTopic process
The literature review contrasts manual, network-based, and probabilistic topic-extraction methods with BERTopic, which captures contextual meaning through semantic embeddings. The BERTopic process represents documents as vectors, clusters them by semantic similarity, and extracts topic representations using c-TF-IDF.
- Overview of Methods in Science Education Literature Research: Science education literature research commonly uses manual content analysis, semantic network analysis, and latent Dirichlet allocation to extract topics or keywords.
- MCA, SNA, and LDA: Manual content analysis codes papers using predefined classifications, enabling in-depth contextual interpretation but requiring substantial time and labor and risking coder subjectivity.
- MCA, SNA, and LDA: Semantic network analysis models keywords as nodes and co-occurrences as links to examine conceptual relationships and influential keywords across large datasets.However, abstracting keywords from surrounding context can reduce semantic nuance.
- MCA, SNA, and LDA: Latent Dirichlet allocation automatically groups frequently co-occurring words into latent topics, supporting large-scale and replicable analysis.Because LDA relies on bag-of-words representations, it does not fully capture semantic relationships among words.
- BERTopic: BERTopic reduces the coding burden of manual content analysis and supports multidimensional analysis by using Sentence-BERT to represent contextual meaning.Unlike semantic network analysis and latent Dirichlet allocation, it captures contextual meaning through transformer-based embeddings.
- Conceptual diagram of the BERTopic process: BERTopic clusters documents by semantic similarity rather than lexical co-occurrence after transforming texts into vector representations.
- Conceptual diagram of the BERTopic process: c-TF-IDF extracts topic representations from each semantic cluster by weighting words according to how strongly they characterize one cluster relative to others.
Data Collection
The study analyzed 2,246 papers from three Korean science education journals, using publication year and English abstracts to identify topics and trends. Author affiliations were categorized by institution type to examine organizational heterogeneity.
- Journal selection: Three journals were selected based on peer review, at least 15 years of publication, publication at least three times annually, and broad disciplinary and school-level coverage.The journals were Journal of the Korean Association for Science Education (JKASE), Journal of Science Education (JSE), and School Science Journal (SSJ).
- Corpus: 2,246 papers were collected, including all SSJ papers published from 2008 onward because records were available from that year.Figure 2 reports annual publication counts overall and by journal.
- Topic data: Publication year and English abstracts were used to identify research topics and their trends.
- Affiliation coding: Author affiliations were grouped into eight institution types to examine organizational heterogeneity.Each author received one affiliation, with university-affiliated teachers enrolled in graduate programs assigned to K-12 schools.
Model Selection
The final BERTopic model was selected from the 11-topic group using statistical and qualitative criteria, with topic structure remaining stable across candidate models. Outlier reduction then reassigned 390 papers while preserving face validity and slightly improving topic coherence and diversity.
- Model selection criteria: 90 candidate models were evaluated using topic frequency, topic coherence, topic diversity, proportion of outliers, and face validity.The criteria combined quantitative metrics with qualitative assessment of whether topics were plausible and meaningful within science education research.
- Model selection criteria: 11-topic models appeared most frequently, and the final model was chosen from this group for its balance of low PO and high TC and TD.Candidate models also had above-average TC and TD and PO values below 40%.
- Model stability: Semantic results were nearly identical across the 11-topic candidates, and topics linked to external conditions appeared consistently, indicating stability across hyperparameter settings.Differences were limited to the boundaries of a few topics, while face validity remained plausible and meaningful.
- Outlier reduction: 390 additional papers were assigned to topics through outlier reduction using c-TF-IDF and SBERT representations, without degrading topic representations.Only papers mapped to the same topic by both representations were reassigned; face validity was unchanged, and TC and TD were slightly higher afterward.
Topic Interpretation
The study interpreted eleven BERTopic-derived topics using representative words and papers, focusing on the concerns each topic reflected. Investigator triangulation supported interpretation reliability and validity while reducing risks from researcher preconceptions and automated methods.
- Topic Interpretation: Eleven topics were named and interpreted using each topic’s top ten representative words and top ten representative papers, totaling 110 papers.The study treated “words” as N-grams for brevity.
- Topic Interpretation: The interpretations identified whether topics centered on policy, curriculum, school-level practice, or subject content.
- Topic Interpretation: Investigator triangulation strengthened interpretation reliability and validity by having science education experts repeatedly review and discuss the results.This approach addressed the risk that a single researcher’s prior knowledge or preconceptions could produce over-interpretation, while recognizing that automated topic modeling cannot replace researcher insight.
Topics and Their Interpretations
BERTopic extracted 11 latent topics from Korean science education publications and organized them into three broad categories based on their prominent orientations. Topic labels were derived from representative words and papers, while category assignments were interpretive and nonexclusive.
- Topic extraction: BERTopic extracted 11 latent topics from the selected Korean science education publications.The topics are presented in Table 1.
- Topic interpretation: Topics were labeled using their top ten representative words and papers to provide an interpretable summary.Representative words appear in Fig. 4, and representative papers are listed in Supplementary Information 2.
- Topic categories: The topics were grouped into sociopolitical, subject-specific, and student-related categories.These categories reflect each topic’s most prominent orientation.
- Topic categories: The three categories are not mutually exclusive because topics can encompass multiple interests and fit more than one group.Topics within the same group may also exhibit different characteristics.
Topic word c-TF-IDF scores
The eleven topics are grouped into sociopolitical, subject-specific, and student-related areas, with university and college authors dominating all topics. Institutional participation varies by topic, especially for educational research institutes, K-12 schools, and other organizations.
- Subject-specific topics: Subject-specific topics follow disciplinary content and inquiry traditions, comprising physics and chemistry, astronomy, biology, and ecology education.The physics–chemistry grouping plausibly reflects overlapping content such as thermodynamics.
- Student-related topics: Student-related topics examine how learners think, feel, reason, and engage in science, including achievement, self-efficacy, motivation, and scientific argumentation.Topic 2 also includes some work on STEM career motivation.
- Institutional affiliations: University and college authors formed the largest group across all eleven topics, ranging from 66.2% (Astronomy Education) to 88.0% (Scientific Argumentation).Educational research institutes published more in Topic 0 (n = 101) and Topic 2 (n = 21), with KICE particularly active.
- Institutional affiliations: K-12 schools accounted for the highest institutional share in Topic 5 (Astronomy Education) at 30.8%, while other institutions reached around 10% in Topic 7.The school share reflects teacher involvement.
Affiliations of all authors by institution type
Topic trends were analyzed across 1,873 papers from 2008 to 2025 and tested against selected external indicators. Significant temporal trends appeared in Topics 4, 7, and 9, while Topic 7 increased without associating with government digital-project counts.
- Topic Trends and Their Interpretations: Trend analysis covered 1,873 papers assigned to topics from 2008 to 2025, excluding outlier papers classified as Topic −1.Topic proportions were calculated as each topic’s annual paper count divided by all papers assigned to topics that year.
- Topic Trends and Their Interpretations: Topics 0, 4, 7, and 8 were identified as candidates for links with specific external indicators.The procedure specified indicators in advance based on representative words and papers, including gifted enrollment for Topic 4 and government STEAM projects for Topic 8.
- Topic Trends and Their Interpretations: The Mann-Kendall test detected significant monotonic trends in Topic 4 (τ = −.60, p = .001), Topic 7 (τ = .66, p < .001), and Topic 9 (τ = −.39, p = .028).The remaining topics showed no monotonic trend, although Topic 0 followed a periodic pattern not captured by the test.
- Topic Trends and Their Interpretations: External-indicator analyses covered 2012 to 2025 (n = 14) because science-gifted enrollment and project-count data began in 2012.Permutation tests were used because the analyses involved few yearly observations.
- Topic Trends and Their Interpretations: Topic 7 increased over time but showed no association with government digital-related project counts.Topic 7 had been expected to increase and to follow those projects; only the increase was confirmed.
Topics with Time-Series Trends
Topic trends varied: teacher professionalism and curriculum implementation followed curriculum-revision cycles, gifted and creativity-related topics declined, STEAM tracked government support, and digital-based science education showed weaker external association. These temporal relationships indicate association rather than causation.
- Science Teacher Professionalism and Curriculum Implementation: Topic 0 followed national curriculum-revision cycles, with M = 5.1 %p, n = 6 during introduction periods versus M = −2.7 %p, n = 8 afterward.The difference was significant: ΔM = 7.8 %p, p = .031; U = 39, p = .030, Cliff’s δ = .63.
- Science Teacher Professionalism and Curriculum Implementation: Topic 0 research increased approximately two to three years after curriculum announcements, despite a pronounced fluctuation between 2021 and 2023.The fluctuation may be associated with disruptions related to the COVID-19 pandemic.
- Interpretive Limitation: These analyses identify temporal correlations with selected external data but do not establish causal inference or directly reveal researchers’ motivations.The interpretations therefore indicate relative policy responsiveness rather than causation.
Trends for selected topics (colored lines) compared with external data (gray bars)
The figure compares annual publication proportions for selected topics with curriculum periods and external indicators. Gray bars show science-gifted student numbers, STEAM projects, and digital-related projects, while topic numbers match Table 1.
- Figure conventions: Annual trend numbers indicate each topic’s proportion of publications per year.The figure uses these proportions to display topic trends over time.
- Figure conventions: Light gray shades mark the introduction periods of the 2009, 2015, and 2022 revised curricula, while dark gray shades mark subsequent periods.The curriculum periods are labeled Ⅰ, Ⅱ, and Ⅲ, respectively.
- External indicators: Gray bars represent the number of science-gifted students, STEAM projects, and digital-related projects.These external indicators appear in panels (b), (c), and (d), respectively.
- Figure conventions: Topic numbers in the figure correspond to those in Table 1.This correspondence links the plotted trends to the paper’s topic numbering.
Topics without Time-Series Trends
Subject-specific and student-related topics maintained relatively consistent publication patterns, reflecting stronger grounding in established academic agendas than in sociopolitical conditions. Ecology education may be an emerging exception, but its recent increase is too short to verify statistically.
- Subject-specific topics: Subject-specific topics remained relatively stable because disciplinary communities and Korea’s subject-based teacher-education structure provide a strong academic foundation.These topics include physics, chemistry, biology, astronomy, and ecology education.
- Subject-specific topics: Ecology education gradually increased in recent years, possibly reflecting growing attention to climate change and ecological transformation in national educational goals.The increase appeared after around 2022, but the trend is too short for statistical verification.
- Student-related topics: Student-related topics were generally less associated with external factors, with achievement, self-efficacy, and motivation research supported by existing quantitative datasets.Scientific argumentation instead showed a notable COVID-19-period decline, possibly because classroom-based studies became harder to conduct.
- Student-related topics: Established academic concerns explain why achievement and motivation and scientific argumentation form part of the stable core of science education research.Their publication trends were relatively constant despite their practical or societal implications.