<![CDATA[Purpose: This study maps the scientific landscape of pedagogical agents in mathematics education (1999–2025) to characterize its developmental trajectory, key actors and intellectual foundations, and thematic structure, and to locate mathematics learning within the broader pedagogical-agents paradigm. Methods: A corpus of 133 Scopus-indexed documents was analyzed with Biblioshiny and VOSviewer, combining performance analysis (production, citation impact, sources, authors, affiliations, countries) and science mapping (keyword co-occurrence at a minimum threshold of three, thematic mapping, and overlay visualization). Findings: The field shows an annual growth rate of 11.25%, far above the global-science baseline of 4%. Knowledge production concentrates in three institutional clusters (Lund University, Carnegie Mellon, UMass Amherst), with international co-authorship at 19.55%, below the cross-disciplinary average. Pedagogical agents anchor the field, with an intellectual base in cognitive principles of learning-by-teaching, while Large Language Models emerged as a converging keyword after 2022. Author-keyword co-occurrence resolved eight clusters dominated by general instructional-technology and AI motor themes, with mathematics learning fragmented across three clusters in a marginal, basic-theme position an application context rather than an epistemic core. Research Implications: Findings highlight open positioning opportunities for new entrants, guide developers integrating LLMs into established pedagogical-agent architectures, and signal to policymakers the need for broader institutional participation. Originality: This is the first bibliometric mapping specifically targeting the pedagogical-agents and mathematics-education intersection, documenting continuity in the paradigm shift from classical to LLM-based agents. As the corpus is drawn from a single database, it should be read as a Scopus-bounded, English-leaning view rather than an exhaustive census.]]>
Copyrights © 2026
