Suspicious Behavior Detection On Social Users

With the rapid development of Internet,social platform has become an important role in our daily life.However,the number of the spam users and Internet mercenaries also increases rapidly.A lot of false accounts and fraud accounts releasing or advertising some malicious information in different social platforms gives great damage and huge threaten to those of normal users in social platforms,which influences the healthy development of social network.In this case,it's important and significant for social platforms to detect abnormal users by analysis the data of users' behavior.Firstly,we propose an evaluation index to measure the degree of the suspicion of the synchronous behaviors.When these data are stored with tensor,the dense sub-tensors are responsible for these synchronous behaviors.The evaluation is based on possion distribution,which is suitable for some Axioms and is better than traditional evaluation index.Secondly,we propose a method on dense blocks detection based on greedy algorithm which finds the block with the highest metric in tensor.And we combine the method with binary space partition tree to find some dense blocks in static tensor.By comparing key values between child's and father's nodes,we judge whether the binary tree grows.When the binary tree stop growing,all the child nodes are dense blocks.End condition of binary tree growing is mathematically proven.Finally,experiments on both synthesis data and real-world data show efficiency of the method.Lastly,we propose a methond on detecting dense blocks in tensor stream.This paper presents a improved greedy algorithm with the new sorting rule in the dense block detection.And this method can find the range of index in which the events have changed by calculation formula quickly.Then the method resort indexes and judge whether we need to detect the dense block in the period.Thus the paper presents the method to detect block in tensor stream with timestamp.Finally,experiments on real-world data show the efficiency of method.