A Clustering Based Feature Selection Approach to Detect Spam in Social Networks

In recent years, online social networks (OSNs) have been expanded with a lot of facilities and many users and enthusiasts have joined to OSNs. On the other hand, the proportion of low-value content such as spam is rapidly growing and releasing in the OSNs. Sometimes the spam advertising purposes, commercial purposes or spreading lies in the different mailing lists are placed and shipped in bulk to send for social network users. Spams not only damage the interests of users, usage time and bandwidth, but also are a threat to productivity, reliability and security of the network. In this paper, we present an online spam filtering system that can be deployed as a component of the OSN platform to inspect message generated by users in real time. Our filtering method is working on the basis of different features such as like, replay, hash tag, followers, and the existing URLs in the posts of Facebook social network. We employ three clustering algorithms for this purpose and we also use naïve Bayes and decision tree to detect spam from non-spam. We evaluate the system using 2000 wall posts collected from Facebook.