LBSNShield: Malicious Account Detection in Location-Based Social Networks
Xuan Yuan, Yang Chen, Huiying Li, Pan Hui, Lei Shi · 2016
Given the popularity of GPS-enabled smart devices, location-based social networks (LBSNs) have attracted numerous users around the world. The openness of LBSN platforms has also made themselves the targets of malicious attack-ers. In LBSNs, attackers can register a number of fake iden-tities and let them post spam reviews or fake check-ins. Therefore, discovering and blocking the malicious accounts are vital for the experience of legitimate users. In this pa-per, we investigated how to accurately detect malicious ac-counts in LBSNs. We collected rich user data from a pop-ular LBSN in China, so-called Dianping. We then built a crowdsourcing based annotation platform to mark legitimate and malicious accounts. By examining the annotated data set, we selected a number of key features to distinguish between these two types of accounts. Based on these fea-tures, we built LBSNShield, a machine learning-based mali-cious account detection system. According to our extensive evaluation, our system can achieve an F1-score of 0.89.