Blacklist Creation for Detecting Fake Accounts on Twitter

Abstract

Social networking sites such as Twitter, Facebook, Weibo, etc. are extremely mainstream today. Also, the greater part of the malicious users utilize these sites to persuade legitimate users for different purposes, for example, to promote their products item, to enter their spam links, to stigmatize other persons, etc. An ever increasing number of users utilize these social networking sites and fake accounts on these destinations are turned into a major issue. In this paper, fake accounts are detected using blacklist instead of traditional spam words list. Blacklist is created using topic modeling approach and keyword extraction approach. We evaluate our blacklist based approach on 1KS-10KN dataset and Social Honeypot dataset and compared the accuracy with the traditional spam words list based approach. Diverse ensemble creation by oppositional relabeling of artificial training examples, a meta-learner classifier is applied for classifying fake accounts on Twitter from legitimate accounts. Our approach achieves 95.4% accuracy and true positive rate is 0.95.

Keywords

• Topic extraction approach
• keyword extraction approach
• DECORATE