Fake Follower Market Customer Detection based on Geographical Distance Information

Abstract

The reputation of social media such as Twitter, Facebook, and Instagram now regard as one persons power in real-world. The person who has more friends or followers can influence more individuals. So the influence of users is associated with the number of friends or followers. On the demand of increasing social power, an underground market has emerged where a customer can buy fake followers. The one who purchase fake followers acts vigorously in online social network. Thus, it is hard to distinguish customer from celebrity or cyberstar. Nevertheless, there are unique characteristics of legitimate users that customers or fake followers cannot manipulate such as a small-world property. The small-world property is mainly qualified by the shortest-path and clustering coefficient. In the small-world network, most people are linked by short chains. Existing work has largely focused on extracting relationship features such as indegree, outdegree, status, hub, or authority. Even though these research explored the relationship features to classify abnormal users of fake follower markets, research that utilize the small-world property to detect abnormal users is not studied.

In this work, we propose a model that adapt the small-world property. Specifically, we study the geographical distance for 1hop-directional links using nodes geographical location to verify whether a social graph has the small-world property or not. Motivated by the difference of distance ratio for 1hop directional links, we propose a method which is designed to generate 1hop link distance ratio and classify a node as a customer or not. Experimental results on real-world Twitter dataset demonstrates that the proposed method achieves higher performance than existing models.