Detecting Tag Spam in Social Tagging Systems with Collaborative Knowledge

Social tagging systems allow collaborative users to annotate shared resources with tags. Since they rely on user contributed content, social tagging systems are vulnerable to spam annotations, which are generated by malicious users to mislead or confuse legitimate users. Thus, mechanisms for spam detection need to be developed to combat the flexible strategies of spammers for the success of social tagging systems. Since annotations are lack of relevant feature, the classical method of training classifier to detect spam is hard to implement. However, with their collaborative nature, knowledge on the tagging scheme do exists in the way numerous participants annotating resources with tags. In this paper, we propose a simple but remarkably effective approach for detecting tag spam in social tagging systems with collaborative knowledge. We harness the wisdom of crowds to discover the knowledge on what should be high quality annotations for resources. This knowledge is then used to tell spam posts from the legitimate ones. A distinct feature of our approach is that, it can be easily extended for user level spam detection and can do well in both levels. The proposed approach is evaluated on data set collected from real-world system. Experimental results show a convincing performance of proposed approach.