Adopting Community Features to Detect Social Spammers

Recent analysis of social network has gained significant attention due to the success of online social networking websites. One of the common problems in social networks is social spammers who disseminate irrelevant information among legitimate users. The problem of spammer detection has been explored in many previous studies. They have mainly relied on network topological features such as in/out degrees, clustering coefficient, etc. whereas in reality, spammers add secondary accounts which are controlled by them to mimic the behavior of normal users. So, spammer detection models which only consider topological features merely offer mediocre performance. So, in this paper we aim to overcome drawbacks of previous models by proposing a spammer detection model which uses a strong community detection method to extract community-based features along with the other features. Also, we apply a feature selection approach to select appropriate features to reduce data and computation, and to enhance generalization. Therefore, by using community-based features which reduces imitation of spammers from normal users, the presented model provides fairly better performance compared to the existing approaches.