Detecting Various Types of Malicious Users at One Sitting in Online Social Networks

Mou, Guanyi

Etd

Detecting Various Types of Malicious Users at One Sitting in Online Social Networks

Public

Online social networks (OSNs) have long been suffering from various types of malicious users such as spammers and bots. Over several years, researchers proposed multiple approaches to identify different types of them toward low- ering their impact into the OSNs. However, their strategies mostly focused on some specific types of malicious users (e.g., spammers, bots), or they less paid attention to newly emerging malicious users. To overcome the limitation of the prior work, in this study, we proposed a novel method to detect various types of malicious users at one sitting. In particular, we (i) combine publicly available Twitter user datasets and categorize these accounts into two groups (e.g., legitimate account, and malicious account); and (ii) propose a robust deep learning framework which jointly learns various features and detects malicious accounts. Our experimental results show that our proposed mod- els outperform stat-of-the-art baselines, effectively detecting various types of malicious users at one sitting. Under the training data reduction scenario, our models consistently achieve high accuracy. Our source code and dataset are available at an anonymized URL.

Creator