BERTopic-Based Topic Modeling and Knowledge Graph Framework for Semantic Mapping of the Scoliosis Literature

This study proposes a semantic framework that integrates topic modeling and knowledge graph approaches to comprehensively analyze the conceptual and structural evolution of scoliosis research over the past twenty-five years. A large compilation of 31,836 English abstracts published between 2000 and 2025 from the Scopus database was used. In the first stage, latent research themes in the literature were identified using the BERTopic method based on contextual language representations. Subsequently, a topic-sensitive knowledge graph was constructed using semantic relationships automatically extracted from the abstracts, thereby bringing together latent thematic structures and explicit relational knowledge. Semantic similarities between topics were analyzed using embedding-based metrics, while temporal analyses revealed the dynamics of research theme intensity and diversification. Findings indicate that scoliosis research is organized around a central core of clinical and methodological themes, alongside more specialized and interdisciplinary themes that emerged over time. Sample-based validation using a relation-wise stratified protocol (50 triplets per relation type), combined with automated confidence-based pre-labeling and manual adjudication, demonstrated moderate-to-high precision across relation categories. By combining machine learning with knowledge representation approaches, the study offers an innovative method for the scalable and interpretable analysis of biomedical literature.