Semiautomated text analytics for qualitative data synthesis

Abstract

Approaches to synthesizing qualitative data have, to date, largely focused on integrating the findings from published reports. However, developments in text mining software offer the potential for efficient analysis of large pooled primary qualitative datasets. This case study aimed to (a) provide a step‐by‐step guide to using one software application, Leximancer, and (b) interrogate opportunities and limitations of the software for qualitative data synthesis. We applied Leximancer v4.5 to a pool of five qualitative, UK‐based studies on transportation such as walking, cycling, and driving, and displayed the findings of the automated content analysis as intertopic distance maps. Leximancer enabled us to “zoom out” to familiarize ourselves with, and gain a broad perspective of, the pooled data. It indicated which studies clustered around dominant topics such as “people.” The software also enabled us to “zoom in” to narrow the perspective to specific subgroups and lines of enquiry. For example, “people” featured in men's and women's narratives but were talked about differently, with men mentioning “kids” and “old,” whereas women mentioned “things” and “stuff.” The approach provided us with a fresh lens for the initial inductive step in the analysis process and could guide further exploration. The limitations of using Leximancer were the substantial data preparation time involved and the contextual knowledge required from the researcher to turn lines of inquiry into meaningful insights. In summary, Leximancer is a useful tool for contributing to qualitative data synthesis, facilitating comprehensive and transparent data coding but can only inform, not replace, researcher‐led interpretive work.

Gender and active travel: a qualitative data synthesis informed by machine learning

“…This was applied via the text analytics software tool, Leximancer (Version 4.51). The detailed rationale for applying Leximancer in this study and a step-by-step guide for doing so are described elsewhere [30]. In short, Leximancer applies a form of text mining that mostly utilises a statistical unsupervised machine learning approach.…”

“…It is important to note that there is a disconnect between the language used within the fields of text analytics and traditional qualitative analysis [30]. These differences are defined in Table 2.…”

“…The only information that we gave the software was the gender of the respondent and the original study of each transcript. The six steps involved in this first stage of the analysis were conducted by one member of the research team (EH) and can be found in the Additional file 1: Table A, and discussed in greater detail in a separate publication [30].…”

“…The aim of applying a machine learning approach to our data was to uncover networks or patterns that had not emerged from the original and more traditional forms of qualitative data analysis of the individual datasets. However, despite the utility of Leximancer in these early stages of concept identification and data coding, researcher-led interpretative work remained an essential component of the analysis and researcher insight fundamental to developing meaningful outputs [30].…”

What are the key themes associated with the positive learning experience in MOOCs? An empirical investigation of learners’ ratings and reviews

“…These logs provided valuable information that guided the authors through the text. However, Leximancer only provides a starting point to the analysis of a large amount of text data and cannot replace researchers' judgement, inference, and interpretation (Haynes et al, 2019).…”