Scientometrics Computer Science Applications

Identifying emerging research fields through semantic enhanced topic modeling: integrating BERTopic and large language models for knowledge structure analysis

Zhang, Yi, Song, Dacheng, Zhang, Ningning, Wen, Ke, Zhang, Yifeng, Lin, Zefu
Springer Nature · 2026-11-19 · DOI 10.1007/s11192-026-05893-7

Abstract

Understanding how new research fields emerge and evolve is fundamental to scientometrics and science policy. Traditional topic modeling methods have been widely applied to detect research trends and hotspots, yet their reliance on statistical co-occurrence limits the ability to capture deep semantic relationships and interdisciplinary linkages. To address this challenge, this study proposes a semantic-enhanced topic modeling framework that integrates BERTopic with Large Language Models (LLMs) for improved identification and interpretation of emerging research fields. BERTopic is employed to extract latent research topics, while LLMs enhance semantic coherence evaluation and guide iterative refinement of topic representations. The proposed approach constructs a hierarchical “field-topic-subtopic” knowledge structure, offering a fine-grained view of scientific domain evolution. Experiments using representative datasets demonstrate that this framework effectively identifies interdisciplinary frontiers and reveals the structural dynamics of emerging fields. This study contributes to scientometric research by introducing a semantic-based analytical paradigm that bridges quantitative topic modeling with interpretive understanding of scientific development, thereby supporting more informed research evaluation and strategic decision-making in science policy.

Read at publisher →