Artificial Intelligence in Science Education: A Data-Driven Mapping of Research Trends, Knowledge Structure, and Collaboration Networks

Abstract

This study presents a bibliometric mapping of the global landscape of artificial intelligence (AI) research in science education over the 2020–2026 period. A corpus of 140 articles was retrieved from the Web of Science Core Collection following PRISMA 2020 guidelines and analyzed using Bibliometrix, VOSviewer, and Biblioshiny. Findings reveal that the field is in an exploratory growth phase, with an annual growth rate of 40.42%; the year 2023 emerged as a pivotal milestone in both publication output and citation impact. Geographically, the United States (n=109), China (n=75), and Germany (n=46) dominate production, while Africa and Latin America remain virtually absent—a pattern indicative of geographic representational inequality rooted in database bias, language barriers, and funding disparities. Bradford's law identified only three core journals, with the Journal of Chemical Education's inclusion reflecting the disciplinary distinctiveness of science education. Intellectually, Cooper (2023) occupies a seminal position with 676 citations; ArXiv's second-place ranking in local citation analysis reveals that science education researchers draw directly from machine learning literature. Thematic mapping shows "artificial intelligence" and "education" as motor themes, while "chatgpt" and "generative ai" remain widespread yet theoretically underdeveloped. The 2026 thematic shift toward teacher and student themes signals a transition from a tool-centered to a human-centered research agenda.