Rumor Detection in Social Media

Rumor is a kind of message that spreads from person to person by word of mouth. It can be a lie, but also information or even a fact that is not yet confirmed (unverified). When a rumor is spread it can become widely accepted as true, although it may later be revealed to be false. Rumors can have serious consequences, such as encouraging people to break the law or causing panic in an emergency situation. Some rumors are easier to detect than others, but it is still difficult to keep up with the volume of circulating rumors. This is especially challenging when people use social media to share rumors, which can be spread very rapidly.

A large body of research exists on rumor, ranging from psychological studies to computational analysis. One early study by Gordon Allport focused on the role of anxiety in rumors. He found that anxiety (situational or personality) and ambiguity are key ingredients to the creation of rumors. Rumors that lift people’s anxieties, or whose content is of high importance to individuals, are more likely to be spread. Ambiguity is a common problem with rumors because people are not sure what is really happening.

The rise of social media has created new challenges for journalists and other people who need to disseminate accurate information. As a result, there is growing interest in automatic methods for identifying rumors in social media before they can cause damage. Many of these approaches focus on detecting linguistic cues that indicate uncertainty, and they are often combined with metadata about the sender such as friend-follower ratio or verification status.

In addition to retweeting debunking tweets, it is also possible for people to write original debunking messages themselves. We studied these Twitter users’ tweets in the aftermath of five major rumors associated with natural disasters during Hurricane Sandy in 2012. We identified three types of original debunking strategies: sharing media content with followers that denies a rumor, addressing the rumor spreader directly by @-mentioning them, and making an unsupported claim to one’s followers that a rumor is false.

The results show that the type of strategy employed has a significant effect on the effectiveness of debunking a rumor. In particular, it is important to provide evidence that a rumor is not true when it is first shared. This is shown by the higher level of support and denial that we observed for debunking tweets that employ evidence-building strategies such as claiming first-hand experience, providing a link to an accessible source, or using reasoning to convince the reader.