Spam Detection in Social Bookmarking Systems Using Tag Scores and Selective Evaluation

In this paper we study spam detection in social bookmarking systems. We analyze posts and develop features to find spammers from users by focusing on analysis of tags. In calculating tag scores, we carefully select tags to give different weights depending on how much spammer and non-spammers like them. To further enhance performance, we propose features based on semantic similarity, tagging tendency, and quantitative tag information and combine them with the tag features. We evaluate the features using the well-known dataset and see they work in detecting spammers. In addition, we do experiments to see how our proposed features work in different environments, namely, different number of posts per user, different classifiers employed, and different levels of feature computation.