In the age of social media, rumors are spreading faster than ever. In the era of Twitter, Facebook, and YouTube, everyone can both create and spread information. However, some of this information is false and can cause great panic in society. Rumors have also become more difficult to dispel.
The concept of a rumor has a long history in human culture, but modern scholarly research on rumors dates back to the work of German social psychologist William Stern in 1902. Stern experimented with a chain of people who passed stories from mouth to ear without having the right to repeat or explain them. He found that the stories were shortened and changed as they moved from person to person. Stern’s experiments led him to formulate the modern scholarly definition of a rumor: “any information which, though unsubstantiated, has enough of the appearance of truth to induce belief.”
Researchers have studied the process of identifying rumors in social media using several approaches. Most methods focus on identifying deceptive content, such as linguistic cues (Kwon et al., 2013) and metadata about the sender (e.g., friend-follower ratio or verification status of Twitter accounts) (Castillo et al., 2014).
While these methods have shown some success, they are not foolproof. Even the best automated systems cannot accurately identify all rumors. Some rumors are too vague or too fast to be identified with these methods. Furthermore, the coding of such large data sets is time-consuming and subject to coder bias.
In this article, we propose a new method for analyzing the process of dispelling rumors in social media. Our method takes into account the personality of each individual. We classify individuals based on their personalities: some are radical and believe what they hear; others are steady and calm and are more likely to contemplate and seek confirmation before believing or spreading a rumor. We find that individuals who are in the latter category are more likely to be responsible for dispelling a rumor.
We apply our methodology to a dataset containing tweets about the G20 summit in Hamburg, Germany. We compare the occurrence of five different false rumors to see how each one is countered in real-time. The results show that rumors that are debunked early and vehemently by professional media and police departments have the lowest likelihood of circulating. We also analyze the conversational characteristics surrounding a rumor in order to understand what factors influence its debunking.