Abnormal Account Detection On Social Platform Based On Main Characteristics

With the rapid development of the Internet since the 21 st century,social network has provided a lot of convenience for work,study and entertainment.People are increasingly dependent on it,and now it has become an important part of people's daily life.Precisely because of the irreplaceable superiority of social platforms in disseminating information,criminals have thus updated the main methods of social engineering attack.Early years they use Trojan or virus to attack,and nowadays they create fake accounts or steal normal user accounts to spread bad information,such as spam ads,phishing websites and pornographic information.This kind of malicious behavior has a wide coverage,strong pertinence and it is difficult to prevent,which has a great impact on the user experience and property security of social platforms.Therefore,the anomaly detection technology of social platform accounts has been the focus of social engineering experts and social platform developers in recent years.From the perspective of social engineering,this paper comprehensively analyzed the academic achievements in related fields at home and abroad,took Twitter users as the research object,calculated the changes of interest of the users according to the tweets posted by them,and then judged whether there was any abnormality in the account.First of all,we summarize the scheme of existing achievements,finding that some features are easy to be imitated by criminals.To solve this problem,Ho HD(Hurst of Hobby Distribution)is constructed based on the Latent Dirichlet Allocation(implicit Dirichlet Distribution model)subject model of LDA.Secondly,the stability of users' interest distribution over time is analyzed,and the stability is quantified by Hurst index.As an account feature,it was sent to the classifier to be learned,and the abnormal account detection effect of the feature was evaluated under different parameters.Compared with existing studies,the detection accuracy and recall rate of abnormal accounts have been improved.Finally,a compound feature selection method based on min-redundancy maxrelevance(m RMR)is proposed.The features proposed in this paper are added to the traditional features for comparison and combination to determine their importance.