Analyzing textual (Mis)information shared in whatsApp groups

Abstract

Whatsapp is a messenger app that is currently very popular around the world. With a user-friendly interface, it allows people to instantaneously exchange messages in a very intuitive and fluid way. The app also allows people to interact using group chats, sharing messages, videos, audios, and images. These groups can also be a fertile ground to spread rumors and misinformation. In this work, we analyzed the messages shared on a number of political-oriented WhatsApp groups, focusing on textual content, as it is the most shared media type. Our study relied on a dataset containing all textual messages shared in those groups during the 2018 Brazilian presidential campaign. We identified the presence of misinformation in the contents of these messages using a dataset of priorly checked misinformation from six Brazilian fact-checking sites. Our study aims at identifying characteristics that distinguish such messages from the other textual messages (with unchecked content). To that end, we analyzed various properties of the textual content (e.g., language usage, main topics and sentiment of message's content) and propagation dynamics of both sets of messages. Our analyses revealed that textual messages with misinformation tend to be concentrated on fewer topics, often carrying words related to the cognitive process of insight, which characterizes chain messages. We also found that their propagation process is much more viral with a distinct behavior: They tend to propagate faster within particular groups but take longer to cross group boundaries.