Computer-Supported Code Discovery Utilizing Topic Modeling and Stepwise Coding

Coding, the process of defining concepts and identifying data, is a critical aspect of data analysis, particularly in quantitative ethnography (QE). While prior methods have used topic modeling—a probabilistic generative model intended to identify possible topics in textual data—for computer-supported code discovery, they have also been criticized for their lack of reasoning transparency and failure to consider the context of the analysis. Thus, we propose a novel method that combines stepwise coding, a qualitative data analysis method for extracting concepts at progressively higher levels of abstraction, with topic modeling. This approach aims to support verifiable code discovery while considering the analysis context. This study examined the proposed method utilizing real interview data on a learning support technology in music education. First, we identified 14 meaningful codes by applying latent Dirichlet allocation (LDA) to the themes and concepts derived from stepwise coding. Second, we compared the topic modeling-based codes with previous human analysis-based codes. Furthermore, we explored the potential for automated code assignment by creating epistemic network analysis (ENA) graphs for topic model-based coding. The results demonstrate that our method captures the analyst’s perspective with a high alignment between the discovered and human analysis-based codes while offering new insights. Our findings contribute to QE research by proposing a transparent and verifiable process for code discovery that strengthens traceability between codes and qualitative data.