Abstract
Social networks present useful tools for communication and information sharing. While these networks have a considerable impact on users daily life, security issues are various such as privacy defects, threats on publishing personal information, spammers and fraudsters. Consequently, motivated by privacy problems in particular the danger of sexual predators, we seek in this work to present a generic model for security policies that must be followed by social networks users based on sexual predators identification. In order to detect those distrustful users, we use text mining techniques to distinguish suspicious conversations using lexical and behavioral features classification. Experiments are conducted comparing between two machine learning algorithms: support vector machines (SVM) and Nave Bayes (NB).