Trust Evaluation Model In Social Networks

Due to the convinence to spread and share information,social networks become the good tool to interaction between people,as well as a good platform for serving people.With the growth of the social networks,many research topics have attracted much interest.For example,to guarantee the security level of the networks is the foundation for social networks to function normally.The establishment of social networks is based on the interaction relationship between users;especially trust between users plays an important role.Without trust guarantee,social networks will suffer from security crisis such as privacy leakage,internet fraud and attack incidents,and bring damages to users.Thus the establishment of trust models between users has been regarded as effective security strategy.Moreover,in social networks,it is necessary to find the most active and influential users to improve information propagation efficiency,and benefit the network marketing.In social network analysis,centrality has been used to measure the importance of nodes.Taking into account the discrepant trustworthiness between users can improve the accuracy of measuring user centrality.Thus,in this thesis,the trust model and its application in centrality measurement in social networks have been investigated.Firstly,the adaptive trust model in social networks is studied.Firstly,direct trust evaluation model is established based on Bayesian system;then,according to the propagation rules of trust in social networks,indirect trust model based on semiring algebra is proposed.Furthermore,the adaptive forgetting scheme is proposed to describe the dynamic of trust.Dynamic adjustment mechanism is adopted to determine the forgetting factor,which is more suitable for the dynamic change rule of trust.Experiment results show that,the adaptive trust model proposed in this thesis can effectively distinguish normal nodes and malicious nodes,which can improve the security of social networks.Meanwhile,the trust updating mechanism based on feedback can improve successful transmission rate in social networks.Secondly,the time-variant stochastic trust model for social networks is investigated.Based on the dynamics of trust on the timeline,the influence of accumulative historical trust information on trust evaluation is taken into account in this thesis.And time-variant trust value model and time-variant trustworthiness threshold model are proposed.Then the warning probability of nodes is derived to alert the untrustworthy level of nodes,thus the safety of the network can be ensured.Furthermore,security threshold is introduced to evaluate the reliability of nodes and adjust the tradeoff between network security level and nodes utilization rate.The consistency of numerical and simulation results verifies the practicability of the TSTE model.Simulation experiment results show that,based on the proposed time-variant trust model,normal nodes can be distinguished from misbehaved nodes,and thus network security can be protected effectively.Furthermore,by analyzing on the tradeoff relationship between network security level and nodes utilization rate,it can be observed that,at each time moment,there is always an inflection point,where both security level and utilization rate can be achieved as high as possible.Therefore in the practical application,the security threshold can be set as the value corresponding to the inflection point.Then the optimal tradeoff between security level and utilization rate can be achieved,and network resource can be used fully at high trust level.Finally,the centrality model for weighted social network is studied.There is difference in interaction relationship between users in social networks,for example the difference of interaction frequency and duration results in the difference of closness and trustworthiness between users.Such social networks with discrepant interaction relationship are regarded as weighted social networks.Since most social networks are weighted,it is necessary to propose centrality model for weighted social networks.Trustworthiness is an important weight on ties between users,and integrating it into centrality measurement can improve the accuracy as well as the network security.In this thesis,the Principle Component Centrality(PCC)for unweighted social networks is firstly extended to weighted networks,and the Trustworthiness-based Weighted Principle Component Centrality(TW-PCC)is proposed.Experiment results show that,TW-PCC model is capable to identify the most important users in the social network accurately,and outperform other weighted centrality algorithms in spreading effectiveness,robustness and tolerance.Furthermore,the most important users of each community in the network can be identified based on TW-PCC model more accurately than other centrality measures.In conclusion,this thesis studies the adaptive trust model,the time-variant trust model,and the application of trust model in key users mining in social networks,and proposes the semiring trust model based on adaptive forgetting scheme,the time-variant stochastic trust evaluation(TSTE)model,and trustworthiness-based weighted principal component centricity(TW-PCC)model.Research on the analysis and application of trust evaluation models in social networks is of important theoretical and application significance.