Research On Abnormal User Group Detection Based On Audit And Reasoning

Social networks are a mapping of social relationships in the real physical world to virtual cyberspace,storing a large amount of user data and user social relationships.User security is directly related to the safe and stable operation of online social networks.At present,malicious users and the correlation of abnormal behaviors are becoming more complicated and hidden.Traditional abnormal detection methods for individual users have been unable to meet the current abnormal user groups detection needs,and the detection methods for abnormal user groups have gradually become a research hotspot.Aiming at the huge amount of behavior data left by social users,an abnormal user group detection method based on auditing and reasoning is proposed.The main tasks are as follows:(1)Discussion and research on methods related to group behavior analysisThe related work of group behavior analysis is summarized.This thesis expounds the tendency and characteristics of group behavior introduces the related concepts of microblog user behavior analysis,and deeply studies the theoretical basis of user behavior analysis algorithm,behavior correlation logic reasoning and other related technologies.(2)User audit model based on attribute measurement and similarity measurement For microblog users in social network,it analyzes user attribute information and self-issued information based on the user audit model.Firstly,the concept of user security degree is proposed,and the security degree threshold is defined to identify malicious users.Secondly,the user attribute measurement algorithm is constructed,the user attribute data is read,the hierarchical weight decision model is used to calculate the attribute weight vector,and the attribute information is analyzed.Finally,the similarity measurement algorithm is constructed,use word segmentation technology to extract keywords from the original blog content,and the Levenshtein Distance is improved.By studying the contents of blog posts,it reflects the preferences and characteristics of users' spontaneous behaviors.Experimental results show that this model can comprehensively analyze the attribute and behavior data of users,and have more accurate and stable performance.(3)Abnormal user group detection model based on correlation probabilistic reasoning.Based on the malicious users detected by the user audit model,the interaction information between users is further considered,and the correlation analysis of abnormal behavior is performed based on probabilistic reasoning.Firstly,the concept of user behavior correlation degree is proposed,which is defined as two parts: user attribute similarity degree and behavior interaction degree.Secondly,the algorithm of attribute similarity measurement and behavioral interaction measurement are constructed to identify abnormal correlation users.Finally,the probabilistic soft logic is used to express the before and after connection of user behavior.A set of reasoning rules suitable for the detection of abnormal correlation users are proposed to complete the probabilistic reasoning of abnormal correlation users,and the abnormal correlation users with the same correlation probability value are determined as the abnormal user groups.Experimental results show that,this model can identify the users correlated with anomalies,make probabilistic inferences on the possible associations,thus achieving higher accuracy and predictability.