Text as big data: Develop codes of practice for rigorous computational text analysis in energy social science

Abstract

Augmenting traditional social science methods with computational analysis is crucial if we are to exploit the vast digital archives of text data that have become available over the past two decades. In this journal, Benites-Lazaro et al. [1] showcase this in an application of topic modeling and other computational methods to an actor-specific examination of changes in policy discourse on ethanol in Brazil and point out methodological promises and challenges. However, their contribution also highlights the need for establishing codes of practice for computational text analysis. In this perspective, we discuss five areas for improvement when treating text as big data in light of guiding principles from computational research – transparency, reproducibility and validation – to facilitate rigorous research practice: (1) full transparency over data collection and corpus construction, (2) comprehensive method descriptions that enable reproducibility by other researchers, (3) application of rigorous model validation procedures, (4) results interpretation based on primary text and clear research design and (5) critical discussion and contextualization of main findings. We conclude that the energy social science community needs to develop codes of practice to build on the promising research within the field of computational text analysis and suggest first steps into this direction.