Research On Reweeting Over Online Social Networks

With the prevalence of computers and the Internet, the spatial limitation of inter-personal communications has been overcome, and online social networks are playingmore and more important roles in our daily life. Consequently, the explosive growthof the amount of information, its rapid propagation, and diverse approaches to infor-mation on the one hand provide us with a lot of convenience, while on the other handbringusnewchallengesonpublicsupervisionandrumorcontrol. Thisdissertationsur-veys the existing studies on information over online social networks and investigatesthe retweeting mechanism using the data from Sina Weibo. The main work and resultsare summarized as follows:1. The existing studies on twitter and blog have shown that users’ behavior ofposting information is periodic. Based on the data collected from Sina Weibo, weconduct a statistical analysis of the amount of tweets during diferent time periods. Theresults illustrate that the behavior of users is also periodic: the amount of tweets has aseven-day periodicity, and there are much more tweets on weekdays than on weekends.Moreover, the peak hours of posting tweets are also diferent between weekdays andweekends, with around11a.m. on weekdays and10p.m. on weekends.2. As one function of Sina Weibo, retweeting is an important way to disseminateinformation. The process of retweeting a piece of tweet can be mapped into a difusionnetwork, in which a node represents a user and a link represents the retweeting rela-tionship. We mainly study the characteristics of difusion networks and fnd that bothof the size and width of difusion networks follow power-law distribution, while thedepth of networks approximately follows exponential distribution. Moreover, we alsofndthatthedurationofretweetingandthetimedelaybetweentwoconsecutiveretweet-ing follow power-law distribution. We divide tweets into six classes according to theevolution of retweeting times. Furthermore, by using an outdegree-indegree method, we classify the difusion network into diferent patterns, and fnd that star graphs andchain graphs are most frequent; there is also a fraction of self-circle, which indicatesthatusermayretweettheirowntweetsandretweetreciprocallybecauseofconvenienceand low cost. In addition, the statistic results show that the retweeting probability isafected by exposures.3. A new information difusion model is proposed based on our empirical study.We suppose that users can be divided into innovative ones and conformable ones inaccordance with their character. The probability that one user retweets the tweets isrelated to both the the level of exposure to the information and the user’s character.Simulations over diferent networks indicate that both of the topological property ofnetworks and proportion of active nodes have efects on information difusion. In par-ticular, a network with high fraction of conformable users might suppress propagationto some extent. The property of initial active users also has impact on difusion: a userwith greater centralities in the sense of degree, betweenness centrality, and eigenvectorcentrality is more likely to spread the initial information to a wide range.