Identifying emotions in texts using the Emoxicon approach to compare right and left-leaning trolls on Twitter

Abstract

Emotion serves an important role in spreading (dis)information online. Although there is extensive research on how the Russian Twitter accounts that purposely spread disinformation ("trolls") behaved during the 2016 U.S. election, the use of emotion by these accounts is less well understood. In this paper, we introduce a new method of analyzing emotion in text called Emoxicon and detail its application to a large set of tweets (n = 420,447) produced by Russian troll accounts. Emoxicon combines an emotion lexicon with Rasch modeling to describe how emotions are used in text. The results from the Emoxicon analysis suggest that the left-wing and right-wing troll accounts used emotion in different ways. Afraid and annoyed were more common emotions among right-wing tweets, while angry and sad were more common among left-wing tweets. However, different results were found when comparing the number of tweets versus the number of accounts. Under all circumstances, happy was the least frequently expressed emotion.