Spam detection for closed Facebook groups

Abstract

Facebook has become a major communication channel for internet users. Unfortunately, with its great popularity and a great number of users, spams are also increasing. A number of Facebook services do not require spam detection, whereas the group usage does. Group users are generally those who are interested in the same topics or purposes. Members usually share the contents of interest in the group. These characteristics enable detection of unwanted posts, referred to as spam that annoys others. It should be noted that some spam may jeopardize the group, for example, by malicious URLs. The objective of this article is to present the design concept for detecting spam in closed groups by using the combination of text features and social features, which comprised 11 features for classifying spam by applying Random Forest machine learning algorithm on 1,200 labeled posts. The result indicated 98% of spam detection efficiency. Additionally, from the feature importance, the number of likes, one of the social features, was found to be the most effective for spam detection.