Epidemiology Inspired Framework for False Information Mitigation in Social Networks

Abstract

Social networking platforms like Facebook and Twitter are used by millions of peoplearound the world to not only share information but also personal opinions about it. Often these information and opinions are unverified, which has caused the problem of spreading of false information, popularly termed Fake News. As social media platforms generate huge volumes of data, computational models for the detection and prevention of false information spreading has gained a lot of attention over the last decade, with most proposed models trying to identify the veracity of the information. Techniques involve extracting features from the information’s propagation path in social networks or from the information content itself. In this thesis we propose a complementary approach to false information mitigation inspired from the domain of Epidemiology. Epidemiology is the field of medicine which deals with the incidence, distribution and control of disease among populations. This dissertation proposes an epidemiology inspired framework where false information is analogous to disease, social network is analogous to population and how likely are people to believe an information endorser is analogous to their vulnerability to disease. In this context we propose four phases that fall in the domain of social network analysis. The first phase is called Vulnerability assessment, where we estimate how likely are nodes and communities to believing false information before an information starts spreading. This is equivalent to assessing the vulnerability (i.e. immunity) of people before infection spreading begins. The second phase is called Identification of infected population, where given the complete spreading paths of information, we identify the false information spreaders from true information spreaders. This is equivalent to identifying infected population after the infection spreading is complete. The third phase is called Risk assessment of population, where given the partial spreading paths of false information, we predict nodes that are most likely to be infected in future. This is equivalent to contact tracing, where we want to identify the exposed population that needs to be quarantined to prevent spreading of the infection. The final phase is called Infection control and prevention where we identify people as false information spreaders, refutation information spreaders or non- spreaders in co-existing false and refutation information networks. This can aid in strategies to target people with refutation information to a) change the role of a false information spreader into a true information spreader (i.e. using refutation information as an antidote) and b) prevent a person from becoming a false information spreader (i.e. using refutation information as a vaccine). Through experiments on real world information spreading networks on Twitter, we showed the effectiveness of our proposed models and confirm our hypothesis that spreading of false information is more sensitive to behavioral properties like trust and credibility than spreading of true information.