Topic modeling as a methodological alternative for qualitative analysis: a case study-based discussion

• In Event: 20420 - Section on Sociology of Culture Roundtables
  In Refereed Roundtable (90 minute): Table 10: Text Analysis

Abstract

With the expanding digital sphere, alternative methods to analyze data have been gaining popularity among social scientists. For cultural sociology, where the meaning remains central to the framework, this approach has been notably challenging. While automating models of text classification presents promising results for characterizing data, their limitations in handling meaning, which is often context-dependent, can obscure the coding process. The size of the dataset and its specificity can add considerable complexity to the task. This paper compares several topic modeling algorithms and one hand-coding method to categorize and analyze data. Drawing from a limited dataset, containing the “About us” section of 408 private art museums from 60 countries, this study discusses the possibilities and limitations of topic modeling as a cultural sociology research tool compared to other coding strategies. While topic modeling has been discussed as an approach for coding data in cultural sociology (DiMaggio, Nag, and Blei 2013), human interpretation as validation is still difficult, given the data sample sizes. By using a limited dataset, this paper seeks to provide insights into the models and their validation according to human interpretation.

Author

• Sara Raquel de Andrade Silva, Universiteit van Amsterdam