Malicious User Detection Based On Trust Computing In Social Networks

Along with the rapid development of Internet technologies, social network itself has become one of most indispensable important applications in people's lives on-line. However, due to the increasing of users, miscellaneous harmful content come into social networks. The existence of alicious users brings very bad consequences to social networks. Ordinary users usually lack of experience and skills, so they are liable to be victims of harassness or frauds which caused by information sent by malicaious users. Social networks often combine with many other Internet applications. Users' accounts of social networks often bind with their payment or financial accounts, mobile phones or personal information. If attackers make use of these, they can conduct severe loss of users' economy and properties or leak of important information. Therefore, design and use of a well-performed model of malicious user detection becomes a very significant work.In view of above situation, starting from the dynamic trust model, the thesis concludes the features and regularities in social networks. On the basis of the rapid change and development trend of current social networks, it proposes a trust model based on interaction information and the trust relation derived from it. It evaluates the user nodes in social networks and adopts up-to-date security defense mechanism, filtering malicious users. This thesis focuses on these assets: First of all, it designes an interaction based float trust model in social networks which makes computation for trust value more tactical. Such model can match each social site better. Secondly, aimed at the problems of huge expenditure on performance and time spending, it proposes an optimization strategy based on time stamp of users' interactions. It allocates weights accouding the time interactions happen. So the computing resources are used in most needed parts such as most intimate relation links. Next, it integrates trust model into malicious user detection system as a core module. Meanwhile some traditional metheds are also used as supplementary. They together make a better accuracy and performance. In the end, serveral simulations are conducted for these three main researches, verifying the rationality and effectness.