[Find updated coverage on this project in the post “The rise and fall of network stars.”]
There has been growing news coverage of entities such as Cambridge Analytica and the mischief that can be done with our personal and sensitive data collected from social networks, such as Facebook and Twitter. As a society, we are starting to understand the many unexpected implications of sharing our private data online. These data can be used not only to “spy” on us, but also to change our opinions, actions, and even our elections.
While Cambridge Analytica demonstrated how third-party applications could collect massive amounts of users’ data, this is only the tip of the privacy iceberg. For years, online social networks have suffered from a wide range of threats to users’ security and privacy. One primary threat is fake profiles, often the root of online social network evils. Recent reports indicate that networks like Facebook and Twitter are infested with tens of millions of fake user profiles. Furthermore, the creators of fake profiles spend considerable effort in making these profiles look as legitimate as possible.
These fake users can collect dozens of personal details about real users and their friends. For example, it’s easy for a fake user to learn your date of birth, workplace, high school name, relationship status, etc. A stranger can scrutinize your personal pictures, obtain your children’s names, and even find your phone number and home address. To make matters worse, there’s often no recovery; you can change your credit card number, but you can’t change your family members’ names.
For the last few years, our research team has studied the threats that jeopardize online social network users, both online and in the real world. As a result, we have developed solutions to better protect online social network users (see video below). Our research team, from both the University of Washington and Ben-Gurion University, specializes in improving online users’ safety by developing tools and algorithms for fake profile identification.
In our most recent research study, published in Springer Social Network Analysis and Mining (SNAM) journal, we developed a machine learning solution to improve social network safety. We constructed a novel anomaly detection classifier to detect fake users. This is a generic unsupervised algorithm that can detect fake profiles by using features extracted from the network structure alone.
Our hypothesis is that a social network user with many improbable links has a higher likelihood of being anomalous — that is, of being a fake user. The algorithm consists of two main iterations. The first iteration creates a link prediction classifier which can estimate, with high accuracy, the probability of a link existing between two users. The second iteration generates a new set of meta-features based on the features created by the link prediction classifier. We used these meta-features and constructed a generic algorithm that not only can detect fake profiles in social networks, but more generally can identify anomalous vertices in different types of real-world networks.
We applied our method on 10 networks of various scales, from a network of several dozen vertices to networks with millions of vertices. In every scenario, our algorithm succeeded in detecting anomalous vertices with lower false positive rates and higher AUCs (Area Under the ROC Curve) compared to other prevalent methods.
Online social networks offer a convenient and fun way to connect with others. Yet real dangers lurk. Our work provides methods to reduce these dangers by identifying and removing fake network users. We all want online social networks to be a safer place, especially for young children and teenagers. Applying machine learning tools to social networks can make all our lives safer
Research team members include:
- Dima Kagan, researcher, Department of Software and Information Systems Engineering, Ben-Gurion University
- Michael Fire, eScience Institute postdoctoral research fellow, Paul G. Allen School of Computer Science & Engineering, University of Washington
- Yuval Elovici, director, Telekom Innovation Laboratories at Ben-Gurion University
“Ben-Gurion University researchers develop algorithm to locate fake users on many social networks”, American Associates, Ben-Gurion University of the Negev.
- Dima Kagan, Yuval Elovici, and Michael Fire, “Generic Anomalous Vertices Detection Utilizing a Link Prediction Algorithm”, Springer Journal of Social Network Analysis and Mining (SNAM), 8(27), 2018.
- Bessi, Alessandro, and Emilio Ferrara. “Social bots distort the 2016 US Presidential election online discussion”, First Monday, 2016.
- Michael Fire, Roy Goldschmidt, and Yuval Elovici, “Online Social Networks Threats and Solutions”, IEEE Communications Surveys & Tutorials, Volume 16, Issue 4, 2014.
- Michael Fire, Dima Kagan, Aviad Elishar, and Yuval Elovici, “Friends or Foe? Fake Profile Identification in Online Social Networks”, Springer Journal of Social Network Analysis and Mining (SNAM), Volume 4, Number 1, 2014.
- Boshmaf, Yazan, Ildar Muslukhov, Konstantin Beznosov, and Matei Ripeanu. “The socialbot network: when bots socialize for fame and money”. In Proceedings of the 27th annual computer security applications conference, pp. 93 – 102. ACM, 2011.