An Evolutionary Analysis of Research Trends in"Assessing Writing"Based on BERTopic Topic Modeling

LI Jinyan; ZHANG Dan; GAO Huimin

doi:10.15926/j.cnki.issn1672-6871.2025.04.011

Abstract：By delving deeply into the research themes of journals and forecasting academic development trends,researchers can more precisely grasp disciplinary directions and keep pace with cutting-edge dynamics.However,journal abstracts,being short texts,present significant challenges to traditional topic modeling methods due to their structured characteristics,high-dimensional sparse vector representations,semantic structural complexity,and data noise.To address this issue,this paper proposes a BERTopic-based model for topic evolution analysis.This model integrates the strengths of pre-trained language models in semantic representation with the structural modeling capabilities of hierarchical clustering algorithms.It also reconstructs the term weighting strategy by introducing a sublinear transformation mechanism for term frequency to optimize traditional term weight calculation methods.This effectively mitigates the interference of high-frequency terms,highlights terms that are critical for distinguishing topics,and significantly enhances the model's topic discriminability and semantic representation capabilities.Using the journal Assessing Writing as the research object,this paper conducts an empirical analysis of research achievements in the field of writing assessment across different periods.By systematically organizing research themes and development directions at each stage,it reveals the dynamic evolutionary patterns of these themes.The experimental results demonstrate that this method can accurately capture changes in research hotspots within the field of writing assessment,clearly reveal its developmental trajectory,and exhibit good practicality and effectiveness in handling short-text data such as journal abstracts.It provides reliable technical support for academic research and trend forecasting in related fields.