Mapping Carbon-Based Sustainability Research Using a Customized BERTopic Framework with TF-IDF–LSA Document Representations

Abstract

This study examines the thematic structure and temporal development of carbon-based sustainability research using a customized BERTopic framework based on TF-IDF–LSA document representations. Covering the period between 2020 and 2026, the study analyzes 58,307 cleaned bibliographic records retrieved from Scopus. Of these records, 55,953 English-language publications containing usable abstracts were included in the topic-modeling corpus. Titles, abstracts, and keywords were combined to construct document-level textual representations. TF-IDF and latent semantic analysis were used to generate reduced document representations, which were subsequently processed through UMAP dimensionality reduction and HDBSCAN density-based clustering. Class-based TF-IDF was then applied to identify the terms that most distinctively represented each cluster. The selected HDBSCAN configuration generated 121 topics and classified 25.8% of the modeling corpus as noise. The findings indicate that carbon-based sustainability research is structured around themes such as the relationship between economic growth and CO₂ emissions, machine-learning-supported carbon prediction, building energy efficiency, net-zero transition, electric mobility, integrated energy systems, low-carbon construction materials, maritime transportation, forest carbon sinks, and agricultural carbon management. The results further demonstrate that the field expanded considerably between 2020 and 2025 and developed an increasingly broad, cross-sectoral thematic structure. By integrating topic assignments with temporal, source, and citation-based indicators, the study provides a large-scale map of the dominant and emerging directions of carbon-oriented sustainability research