The Analysis,Modeling And Prediction For The Structure Of Online Social Network

Online social network(OSN)is a social relationship structure which is formed by multi-agent interactions.As spontaneously formed social structure,OSN contains ample patterns,rules and mechanisms of human behaviors.To discover network structures and explore hidden principles can help us understand human behaviors which will bring potentially commercial values.Thus,studies on OSN have a greater theoretical and practical significance.Using the theories and methods of complex network,we study the evolution and prediction mechanisms of OSN structures.This paper mainly includes two parts: the first part is the studies on the properties and evolution rule of the group structure in QQ group networks and the second part focuses on twitter social network to design a new link prediction algorithm for user following link.The research on the QQ group networks offsets the deficiency of current studies about OSN.So far researchers have generally paid their attentions on individual social relationships,leaving participation in social groups less understood.It is because effective and credible data of collective human behaviors as a group is hard to collect.Through the analysis on the characteristics of QQ group structures,we find two different types of online social groups' growing rules.Further,on the basis of empirical findings,we propose a percolation-like diffusion model to explain the social groups' evolution rule.The model results indicate that the proposed mechanism is an important driven-factor for the growth of real social groups.Moreover,the research work on the link prediction for user following relationship in twitter social networks which covers the shortage of existing link predictors in complex network.Social networks like twitter are interest-driven online community and the posted tweets can explicitly reveal users' personal tastes.While existing link prediction methods in complex network are unable to use text information.In order to make up the shortage,we propose a new local indice of link prediction which can both utilize users' interests and network topology.Experiment results on twitter and sina weibo show that the new algorithm can outperform other traditional link predictors in AUC,precision and reall.