The Research On Key Technology Of Abnormal User Detection In Social Media

With the rapid development of Internet and Web2.0,online social media has penetrated into almost all areas of life including education,medical,entertainment,business et.al.However,while bringing various conveniences to people,online social media has also become the primary target of malicious users who attempt to execute illegal activities and harm other users.Therefore,abnormal user detection technology for such malicious behavior has become one of the key issues in online social media security research.Online social media is dynamic and real-time every day.In previous work,there are few research on anomaly detection in dynamic social media.In addition,these studies only considered non-weighted or undirected networks.The main research content of this paper focuses on abnormal user detection in dynamic social media.Based on the existing abnormal detection methods,this paper proposes two abnormal detection algorithms respectively for abnormal individual user and abnormal group users in social media.For the abnormal individual user detection in social media,the existing algorithm can only be used to analyze undirected weighted graph and cannot solve the problem of abnormal in the case of mapping the user interaction state into directed weighted graph model.Therefore,based on the existing algorithm,this paper proposes an improved anomaly individual user detection algorithm based on the evolutional self-network structure.This algorithm maps a dynamic social media into a set of weighted graph sequences.By comparing the changes of nodes,directed edges and weights of two adjacent graphs in time sequence,the suspicious abnormal nodes sets are filtered out.By analyzing the structural connectivity and intimacy transitivity between nodes,the core network of all nodes in the suspicious abnormal nodes sets is constructed.Finally,the abnormal scores of all suspicious anomaly nodes are calculated by comparing the changes of node's core network in two adjacent graphs.Node with high abnormal scores is considered as abnormal node.The experimental results revealed that the improved algorithm can detect the abnormal individual user in social media effectively and reduce the false positive rate.For the detection of abnormal group users in social media,the existing algorithm can only be used to analyze undirected unweighted graph and cannot solve the the problem of abnormal detection in the case of mapping the user interaction state into an undirected weighted graph model.Therefore,based on the existing algorithm,this paper proposes an improved algorithm for abnormal group users detection based on evolutional relationship.This algorithm maps a dynamic social network into an undirected weighted graph streams.The stream sample is generated by sampling the edges of each graph object in the graph stream for multiple times.The node partitions are generated according to the connected components in the stream samples.All nodes in a connected component represent a partition.The abnormal scores of each graph object is calculated according to the edge generation probability model.Graph object with high abnormal scores is considered as abnormal graph object.The experimental results revealed that the improved algorithm can detect the abnormal group users in social media effectively,reduce the false positive rate and improve the computational efficiency.