Deep Learning Based Attack On Social Authentication System

Abstract

As the rapid development of the Internet, social networks sites, such as Twitter and Facebook, are become more and more popular. These sites require users to provide various kinds of personal data to personalize their services. However, unintentionally disclosed personal information utilized by hackers may pose many security problems. In this paper, we present a deep learning based attack on the social authentication in the Facebook site. Social authentication, known as the fourth-factor authentication, utilizes social interaction information to achieve the identification of users. In the process of account recovery and remote login, the social authentication which requires users to identify names of friends displayed in photos is used as a part of the authentication module. After collecting public information of friends’ photos through web crawler, the attach system uses deep learning networks based on face_recognition library and convolutional neural network (CNN for short) to train the model. The attack is achieved through the successful identification of friends’ names using the trained deep learning model.