Topic models meet discourse analysis: a quantitative tool for a qualitative approach

Abstract

Quantitative text analysis tools have become increasingly popular methods for the operationalization of various types of discourse analysis. However, their application usually remains fairly simple and superficial, and fails to exploit the resources which the digital era holds for discourse analysis to their full extent. This paper discusses the discourse-analytic potential of a more complex and advanced text analysis tool, which is already frequently employed in other approaches to textual analysis, notably topic modelling. We argue that topic modelling promises advances in areas where discourse analysis has traditionally struggled, such as scaling, repetition, and systematization, which go beyond the contributions of simpler frequency and collocation counts. At the same time, it does not violate the epistemological premises and methodological ethos of even the more radical theories of discourse, we will demonstrate. Finally, we present two small case studies to show how topic modellingwhen used with appropriate parameterscan straightforwardly enhance our ability to systematically investigate and interpret discourses in large collections of text.

Grounded reality meets machine learning: A deep-narrative analysis framework for energy policy research

“…In TMs, the meanings of a word are topic-specific and based on the other words that appear in the topics in which it features prominently [25] . This prominence of topics or words is represented through probability scores in TM.…”

“…In TM, the topics themselves obtain their meaning through i) the inter-relationship between words in a cluster of topics; ii) inter-cluster meaning of words between the topics, and iii) through frequent co-occurrences with other topics. Further meaning is derived by linking words to multiple topics, indicating the relationality feature of TM [25] . Besides, GT strengthens the relationality to develop critical insight for policy applications [15] .…”

“…In TM, topics are interpreted as frames, themes, et cetera, but that the most appropriate interpretation depends on how the method is used by the analyst [25] . We expand on this statement using science mapping to examine the commonality and cross-fertilisation between keyword set A and B, respectively.…”

“…This epistemological fit aided in understanding how topic models generate knowledge about the texts and the words in the corpus for fulling the objectives of ‘deep-narrative’ analysis. A similar line of argument was used by Jacobs and Tschotschel [25] for theoretical compatibility validation of their combined application of topic modelling and discourse analysis in qualitative research.…”

“…This may lead to directional biases in its generalised interpretation. A methodological advancement to a similar discourse-analysis through topic modelling was proposed by Jacobs & Tschötschel [25] which provides a critical epistemological background for this study.…”