Rumor Diffusion, Support and Denial in Real Time

Rumor is a piece of information that has not been verified as true or false. It is typically circulated and spread via social media, and can have significant impact on people’s lives. Several real-life incidents of damage and chaos caused by the rapid spreading of a rumor during breaking news events have highlighted the need for automatic rumor detection and verification. However, this is a challenging task because rumours are often temporally and linguistically dynamic, and require real-time collection of reaction as they unfold. Furthermore, rumours are typically based on out-of-vocabulary words (OOV), and pre-trained word embedding models cannot address this challenge.

In addition, there is a tendency for users to support unverified rumours, even when they are later proven false. This may be because of the arousal that these early rumours produce, or their potential for large-scale societal impact. In contrast, tweets denying a rumour do not seem to arouse the same sense of urgency and are rarely retweeted.

This paper studies the dynamics of rumour diffusion, support and denial in real time, by analysing twitter conversation data on nine events involving a rumorous story that was reported as false or true. We collect and annotate thousands of response tweets from users to a source tweet that introduced the rumourous story, revealing how they react and respond to this new information as it spreads. The results show that while rumours are highly contagious, they can be stopped in their tracks when there is an official announcement or evidence dispelling them.

For example, in the case of a single tweet reporting an explosion at the White House, which was quickly debunked, this reduced the diffusion of this rumour. However, for more serious rumours that can have major impacts on the public, such as a message stating that there is a chemical plant explosion in one of China’s provinces which has prompted a massive number of car accidents, it might be impossible to stop this rumour.

Figure 2 shows the distributions of support ratio and certainty ratio by veracity status. The ratio of supporters seems to be relatively stable over time for rumours that are eventually resolved as being either true or false. However, for rumours that remain unverified, the ratio of supporters decreases significantly once they are resolved.

Similarly, the proportion of Twitter users who express certainty in support of a rumour also seems to remain fairly constant irrespective of whether the rumour is ultimately proved true or false. This is probably due to the fact that many users have a desire to persuade others in the truthfulness of the rumour, and as a result are motivated to pass it on.