Topic Modeling in Sociology Using Social Welfare as an Example: Methodological Challenges and the Human Component

Abstract

Considering the dynamically evolving realms of social sciences influenced by network technologies and digital humanities, it is crucial to examine the adequacy of sociological data analysis methodologies in these new conditions. The availability of extensive digitized datasets not only poses a challenge to “classical” analysis methods developed under different circumstances and for different purposes, but also raises the question of whether the traditional demarcation between quantitative and qualitative methods, marked by a clear boundary, remains relevant in the era of Big Data. In this paper, based on topic modeling utilising Latent Dirichlet Allocation (LDA), we argue that quantitative methods (probabilistic statistical models) are not merely complementary or a starting point for qualitative analyses (the standard approach), but, rather, constitute an integral part of them. This thesis is illustrated through a case study involving the identification of themes within a dataset of 17,278 articles published in Web-of-Science-indexed journals between 1992 and 2020, focusing on social welfare. This empirical case study also serves to formulate meta-theoretical observations regarding the “cohesion” of quantitative and qualitative methods in the context of machine learning and natural language processing.