Unpacking Brand Communication on Social Media Through Top-Down and Bottom-Up Text-Mining

This research method case retraces the stages and discusses the practical challenges encountered while conducting a large-scale, text-mining analysis of brand communication on social media during the Covid-19 pandemic. Since research on the topic is scant and theory provides insufficient normative guidance, both top-down, dictionary and rule-based techniques and bottom-up, hashtag network analysis, were employed for serving, respectively, theory testing and theory building purposes. A text-mining design was selected given its ability to assess large volumes of unstructured textual data and to detect patterns that would be otherwise unnoticed by human eyes, to boost the study’s ecological validity by investigating phenomena whose linguistic representations take place spontaneously in natural and unabridged settings, and to explore new phenomena for which conventional dataset are not available. Social media data were retrieved by calling a dedicated Application Programming Interface (API) and prepared for the analysis through several wrangling and corpus-preprocessing tasks. Data analysis encompassed a reiterative process during which each selected technique was thoroughly validated, and results used to adapt the theoretical framework and inform our research questions. Several technical, practical, and ethical concerns are discussed. Overall, the case outlines the advantages and disadvantages of using such a research design, offers best practices to follow, and discourages the blind application of off-the-shelf text-mining services.