Mapping research on teachable agents in mathematics education: A bibliometric analysis of trends, themes, and emerging directions

While teachable agents have been investigated for nearly two decades as a vehicle for the learning-by-teaching paradigm in mathematics education, no prior study has systematically mapped how this body of work is distributed across actors, venues, and thematic clusters, or how it has shifted in response to advances in intelligent tutoring and generative AI. To address this gap, this study employs Scopus-based bibliometric approach to map the intellectual structure, thematic evolution, and publication trends of research on teachable agents in mathematics education. Metadata were obtained through structured Boolean queries and cleaned using OpenRefine to normalise keyword variations (spelling, plural/singular, hyphenation), yielding 60 documents for analysis. The results demonstrate variability in scientific production, with peaks observed in 2011, 2018, and 2025, as well as a concentration of publications in technology- and computer science-oriented venues, particularly Lecture Notes in Computer Science. The actor structure demonstrates the predominance of core research groups and the contribution of specific institutions, with national production concentrated in the United States and Sweden and relatively limited international collaboration. The keyword network forms five clusters, connected by the learning-by-teaching paradigm. There is a shift in themes from the foundations of learning systems and intelligent tutoring towards issues of implementation, scaffolding, personalisation, and the emergence of discourse on generative AI-based agents in the most recent period. These findings provide an empirical basis for future research expanding the context, design, and evaluation of teachable agents in mathematics education.