Peer-reviewed article
Mapping the Intellectual Landscape of Giftedness in Early Childhood Through Comparative Topic Modeling.
Karakaş Mısır S ·
Summary
Written by AI, not yet AICS's view ·Key points
Limitations
Abstract
The present study investigates the semantic structure, dominant themes, and temporal evolution of research on giftedness in early childhood through a comparative topic modeling approach. A final analytic sample (n = 518) of peer-reviewed journal articles indexed in the Scopus and Web of Science databases was analyzed. Three topic modeling methods, Latent Dirichlet Allocation (LDA), Structural Topic Modeling (STM), and BERTopic, were systematically compared using multiple evaluation metrics. BERTopic demonstrated the strongest overall performance, producing approximately 11% higher coherence than STM and approximately 34% higher coherence than LDA. In terms of diversity, it achieved 14% to 17% greater thematic variety and, according to the Gini coefficient, revealed a 58% to 60% more balanced thematic distribution. BERTopic-based analyses identified five major thematic axes: Socio-Linguistic Development and Family Context, Psychometric Intelligence, Identification, and Cognitive Differences, Program Access, Identification, and Educational Equity, Early Academic Skills and Cognitive Development, and Creativity, Higher-Order Thinking, and Enrichment Programs. Thematic mapping and topic similarity analysis were used to examine the semantic structure of the field, while linear regression-based trend analysis over the 1918-2026 publication period showed that family context, socio-linguistic development, and equity-related themes have gained increasing importance over time, whereas psychometric identification largely maintained its central position within the field. These findings indicate that the field is moving toward a more inclusive, semantically grounded, and equity-oriented perspective. However, they should be interpreted in light of the study's reliance on article abstracts, the sensitivity of BERTopic clustering parameters, and the use of linear trend modeling.