MEASURING EMOTION AND MORALITY IN DIGITAL CAMPAIGNS: METHODOLOGICAL CHALLENGES OF AUTOMATED SENTIMENT ANALYSIS IN ELECTIONS

Abstract

The growth of social media has expanded the possibilities of studying political behaviour from large volumes of textual data, but operationalizing concepts such as emotion and morality remains a methodological challenge. This paper asks to what extent automated text analysis techniques, lexical dictionaries among them, can capture moralized and emotionally charged content in digital political communication. The material comprises 2,716 original Twitter posts by Lula, Bolsonaro and Ciro Gomes in the 2022 Brazilian elections. Morality was measured with a moral foundations dictionary, emotional valence with a supervised classifier, and regression models relate these features to the engagement received. The indicators reproduce the patterns of the international literature, with greater diffusion for negative and moralized content. Examining the measures reveals two problems that no conventional test detects. The moral result is carried by a single dictionary category, in which 61% of its occurrences are the words bom and boa. Topic modeling split the corpus by post format before splitting it by subject, and it is this control that eliminates the association of emotion. The paper organizes these findings into a typology and concludes that the validity of a measure depends on the relation between instrument and corpus.