Biterm for spam filtering in short message service text

Abstract

Due to rapid growth in mobile phones usage and reducing cost of sending text messages across mobile networks, short message service has become the most popular communication mode. This move has attracted spammers to mobile networks. Although several machine learning methods have been developed to filter out SMS spam from mobile phone users' inboxes, Short Messaging Service has issues that posse challenges to the use conventional document models that rely on proportion of word distribution. For instance, SMSs suffer from severe sparse context information, which hampers classification of content based on proportion of word distribution. This paper proposes an algorithm that uses biterm topic model (BTM) to model SMS text message. Biterm topic model directly models the generation of word co-occurrence patterns (i.e. biterms)in the whole document. Finally, support vector machine (SVM) was used classification. The algorithm has proved that it can effectively model SMSs for classification using SVM.