G-Fake: Tell Me How It is Shared and I Shall Tell You If It is Fake

Abstract

The propagation of fake news is an increasingly serious concern in social platforms, and designing methods to automatically detect them and limit their spread is an important research challenge. Most existing methods rely on inspecting the content of news to decide on their veracity, but this information is not always available. In this paper, we present G-Fake (Graph-Fake), the first fake-news detection method that is entirely network-based. G-Fake only relies on the sharing history of news items. It does not assume any information on the content of these items (e.g. text or pictures), nor on the trustworthiness of users. In fact, G-Fake does not even require access to the underlying social graph, nor to the interactions between users. Our experimental evaluation conducted on real-world data shows that G-Fake can limit the spread of fake news in the earliest stages of propagation with an accuracy of 96.8%.