Research Of Detecting Abnormal Account In Online Social Network Based On Trust Rank

Online social network is a kind of network service that is based onweb2.0websites. It helps people expanding daily social activity to onlinenetwork, allowing the user to register an account and interact with eachother in the network. Though good behaviors are promoted, thephenomenon still exist that generating spam message through the onlinesocial network account. Because the online social network is open andreal-time, the junk information can spread quickly and widely. As a result,negative events caused by the junk information tend to be worse and worse.Therefore, taking steps to detect and restrict the account which is used togenerate junk information is important to reduce junk information in theonline social network.The major contribution of this dissertation is to introduce the conceptof trust into online social networks. Then a trust evaluation model foronline social network account is proposed. The evaluate outcome can beused to rank the account indicating that which account is more believable.Besides, the information underlying the account relationship is exploited torevise the evaluate outcome. This ordering can be used not only forabnormal account detection in social network, but also as criterion foraccount reliability. The main results are as follows:1) Proposing the trust evaluation model for online social network basedon account characteristics and behavior characteristics. Thedissertation proposes several account characteristic and behaviorcharacteristic which can be used to distinguish the normal account andabnormal account. In this evaluation model, we introduce the Rough Set theory to remove the useless characteristic and bring in a simplesimilarity evaluation algorithm.2) Putting forward the AccountRank algorithm to amend the evaluationoutcome and get more accurate result. It is the phenomenon in theonline social network that the more attention an account attracts, themore believable the account is and the more believable account anaccount contact with, the more believable the account is. Refer to thefamous PageRank algorithm, we proposed a similar algorithm, namedAccoutnRank, to take account of the information hidden in therelationship and interbehavior between accounts.3) Collecting a large amount of data from weibo.com for experiments.Experimental result show that the calculated trust value from ourevaluation model can be used to sort the account, which provides apowerful basis for user to make decision. The abnormal account canbe detected by setting a reasonable threshold.