Structure Analysis And Predictive Modeling In Social Networks

Social network,which contains of complex interactions among social individuals,can reflect a large number of human behavior characteristics.Many studies have shown that most real social networks tend to have some special properties.These properties always reflect different cultural characteristics and help us to understand the develop-ment of society more clearly.In this thesis,we focus on three real-world networks and quantify the significant phenomena of group segregation in the three networks.Quantita-tive analysis shows that,the cross-group edge plays a key role in maintaining the global connectivity of the social networks.Thus,we propose a new link prediction algorithm according to the group characteristics of networks.The problems and results of this thesis are as follows:?1?This work quantifies the phenomenon of group segregation,and reveals the im-portance of cross-group edges in maintaining network connectivity.Through analyzing the visualization of the network topological structures and degree distribution,we find these networks both follow the power-law-like distributions and have obviously segre-gation communications.Then we synthetically analyze the separation coefficient and mixing pattern,and quantify the extent of group segregation.Finally,based on the link percolation dynamics,we show that the corss-group edge plays a critical role in main-taining the global connectivity of the whole network.?2?This work studies the online communication among different religious groups in China.Quantitative analysis shows a high extent of religion segregation in China and reveals important factors for promoting the religious syncretism.We obtain a directed so-cial network from0)4)?7.?88?,which is consisted of 6875 believers in Buddhism,Chris-tianity,Taoism and Islam.Comparative analysis shows that the religion network is highly segregative:the extent of segregation for different religions is far higher than that for dif-ferent races and slightly higher than that for different political parties.In addition,we investigate the few links between different religions and find 46.7%of them are associ-ated with philanthropy.Moreover,Buddhism also plays a key role in promoting cross religious communication.?3?Based on the feature of group segregation and previous researches on link pre-diction,this thesis provides a new link prediction algorithm by taking both the group information and the network structure information into consideration.At the level of group segregation,the likelihood of creating a link between two node4?and5)is consid-ered to be proportional to the closeness between₄)'s group and₅)'s group.At the level of structure information,we find preferential attachment mechanism is significant,since the networks follow the power-law-like distributions.By combining these two factors together,we get a new link prediction algorithm.Experiment results show that the new algorithm out outperforms other benchmark algorithms in both stability and accuracy.