A trust-based approach for security policies enhancement in dynamic social networks

Abstract

The sensitive information revealed in social networks is a real threat menacing users' privacy. Hence, trust is required to enhance the security policies in these networks. Motivated by the privacy problems in particular the danger of sexual predators and the disregard of the dynamic aspect of social networks, users popularity and activity which may affect the precision of the expected results of influential users detection, we aim to present a generic model to improve security policies. To do so, we use text mining techniques to distinguish suspicious conversations using lexical and behavioural features classification and to determine influential users having the maximum trust score value. This trust score is calculated using an algorithm called DynamicInflu inspired by the honey bee's foraging behaviour based on popularity and activity parameters. The experiments are performed on real data and the comparison of the DynamicInflu algorithm with commonly used approaches showed a good performance.