The Detection Of Anomaly Accounts Based On Behavioral Analysis For Social Networks

With the development of the Internet,social networking service(SNS),also called social media or social networking site,has become an important tool for communication in people’s daily life.People from different groups and background use and communicate on the same platform because SNS contains rich information and provides services to users in varied types.Some popular and influential social networking sites attract millions of users to use.SNS provides its users a great convenience in some ways,for example,it reduces the distance of space and time among users.Although SNS broke down the spatial and geographic borders of users,the huge amounts of users’personal information,stored by SNS,has become the target of criminals and third parties.Especially,the anomaly accounts releasing maliciously and advertising information in some social networking sites cause great damage and threaten to those of the users and also to the public.In this situation,SNS needs to recognize and deal with the anomaly accounts by detecting the user behavior data subjectively and regularly,which will play important role and have significance.Based on researching and studying of relevant results of worldwide SNS user behavior analysis,this paper selects Sina Weibo’s data as the data source.The Hidden Markov Model(HMM)was used to model and detect the anomaly accounts,and the statistical analysis can be carried out.First of all,through several demonstrations and comparisons,this paper uses the web crawler technology to collect and process the user behavior data from Sina Weibo,and designs the Weibo information crawling technology framework based on the web page analysis algorithm.Through obtaining URL address in Weibo,establishing the user list,and connecting the accounts automatically,we apply Python language and MySQL database technology for achieving crawling Weibo information.According to the characteristics of user accounts,we analyze the data that we got.And then we select the HMM as the analysis model of social networking site’s user behavior.This paper proposes and builds a model to detect anomaly user behavior in SNS based on the process of hidden Markov.We train our model and get the parameters by using the data collected from the Sina Weibo.Then we use our model to detect the user behavior and determine the Weibo anomaly users,by calculating and judging the max probabilistic paths of the hidden variable states from different observable sequences,and therefore the normal and anomaly users in the data set can be distinguished.We also analyze the statistical characteristics of the test data,and the experiments prove that our model can effectively detect anomaly accounts in Weibo’s data.According to the Weibo user behavior characteristics,we select ghost followers who have the specific characteristics to train the HMM and detect the ghost followers.The experiments show that this model can effectively detect the ghost followers’accounts with certain characteristics in social networking site(Weibo)and also confirm that the correct detection probability of ghost followers’ accounts based on hidden Markov process is greater than the correct detection probability of anomaly accounts in the usual sense.Finally,an anomaly accounts detection system based on hidden Markov process is designed and implemented.This system can detect the anomaly accounts,ghost followers,aggressive following users,aggressive reposting users and aggressive advertising users.The dimensions and parameters of different hidden variables and observable variables,which build up a good foundation for the deep research of anomaly accounts based on hidden Markov process,can be chosen in this system.