Deliverable D5.1 “Software libraries built on Graphos.ml for detection of fake activity in large scale OSNs”

Online Social Networks (OSNs) play an important role in the way that people communicate and consume information. This is mainly because OSNs provide an ideal environment for communication and information acquisition, as users have access to a staggering amount of posts and articles that can share with others in real-time. Unfortunately, OSNs have also become the mechanism for massive campaigns to diffuse false information. In particular, recent reporting has highlighted how OSNs are exploited by powerful actors, potentially even state level, in order to manipulate individuals via targeted disinformation campaigns.

The extensive dissemination of false information in OSNs can pose a major problem, affecting society in extremely worrying ways. For example, false information can hurt the image of a candidate, potentially altering the outcome of an election. During crisis situations (e.g., terrorist attacks, earthquakes, etc.), false information can cause result in wide spread panic and general chaos. False information diffusion in OSNs is achieved via diverse types of users, which typically have various motives. A diverse set of users are involved in the diffusion of false information, some unwittingly, and some with particular motives. For example, terrorist organizations exploit OSNs to deliberately diffuse false information for propaganda purposes. Malicious users might utilize sophisticated automation tools (i.e., bots) or fake accounts that target specific benign users with the goal of influencing ideology. No matter the motivation, however, the effects false information has on society clearly indicate the need for better understanding, measurement, and mitigation of false information in OSNs.

In this deliverable, we provide a taxonomy of the false information ecosystem, we attempt to understand how misinformation is spread and we describe our methodology towards detecting such attempts.

Furthermore, all detection mechanisms have been implemented using state-of-the-art machine and deep learning methods (using keras, tensorflow, Theano, sk-learn)