Fake News Mitigation in Social Networks

Abstract: Fake news is particularly virulent in social networks, where the absence of lters helps itspread quickly. Despite the eorts put forward by the main social network companies, misinformationis still a rampant problem online. One way to limit its diusion is to expose usersto fact-checking content that debunks trending fake news stories with the hope that this willprevent them from sharing misinformation further. This intervention can be carried out by selectinga seed set of users to be exposed to this so-called mitigating content. In order to identifythe best users to select for mitigation purposes, synthetic experiments are performed to simulatethe behavior of the network when the intervention is carried out. These simulations rely on astochastic point process-based model which has been shown to adequately capture the propagationof two competing information campaigns in a network. Due to the overwhelmingly largesizes of the most popular social networks, this model is trained based on real data collectedfrom a smaller sub-network of users who have been carefully chosen. Several algorithms areused to select interesting seed sets of users and their mitigation performances are compared inseveral dierent settings. It is shown that the proposed method of a mitigating intervention inthe network greatly increases the total amount of fact-checking articles over the total amount offake news and that the choice of which users to inuence is signicant as it strongly aects themitigating performances.