Predictive Models for Early Detection of Hoax Spread in Twitter

Nowadays social media are widely used daily to access to news. Indeed, the social media network allows a fast and wide spread of news. Unfortunately these platforms used by millions of people are not immune to misinformation because everyone can be a source of information. Rumors of celebrities death on social media spread very widely in a short time and are hardly verifiable. These kinds of rumors can lead to worrying or stressful situations, and may also have economic or political repercussions. In this work, we have addressed the problem of death hoax diffusion on the social media Twitter. We have collected data related to 25 rumors (false and true) of the death of well-known celebrities on Twitter. Then, we have observed temporal differences and commonalities between true and false rumors in terms of diffusion dynamic, messages and user characteristics. From these empirical observations, we have trained several models to classify early true rumors and hoaxes. We have obtained a rate of correct classification of 0.9 from 20 minutes after the beginning of the diffusion.