Exploring the relationships between emotions in fake news

Abstract

Social media has emerged as one of the main conduits of information for individuals, owing to itsimmediacy and social interactivity, enabling users to share content they find relevant. However, thiswidespread use has given rise to the proliferation of fake news—publications that lack authenticity or areentirely false. In the state of the art, various approaches have been taken to address fake news, includinguser-centric, propagation-centric, interaction-centric, and textuality-centric perspectives. This articlecontributes to the latter by conducting empirical work that seeks to identify fake news and characterize theemotions associated to these fake news in a dataset comprising social media posts. Utilizing the datasetsGossipCop, PolitiFact, and CoronaFake, the objective is twofold: (1) train models for the identification offake news and (2) identify and characterize the emotions associated with fake news. This encompasseseverything from collecting tweets to characterizing emotions. The undertaken work concludes that, onaverage, the algorithms employed achieve an accuracy of 90.2% in identifying fake news. Furthermore, itprovides a characterization of the emotions associated with fake news within the datasets.