Spammer Behavior Analysis And Detection In Online Social Network

In online social networking platforms,users who create and disseminate false information such as frauds,speculations and rumors are called spammers,spam users,or malicious users.The emergence of these spam users has broken the favorable network social environment.In recent years,the problem of spammer detection in online social networks has aroused wide spread concern in today’s academia and industry.Early detection methods mainly rely on text content data to construct user behavior characteristics to capture the abnormal language style of spammers.After that,researchers have conducted a lot of exploration on how to define user behavior characteristics from the perspectives of time,space or rating.However,spammers are becoming more and more adept at disguising and their behaviors are becoming more and more concealed.As time goes by,it is obviously not feasible to obtain and analyze massive amounts of content data in full.Therefore,recent studies have begun to improve the detection method of spam users with the help of complex interaction behaviors in social networks,which do not rely on text content and exist explicitly in social networks.Existing research has not discussed the network interaction relationship in detail,nor has it clarified the mechanism of the difference in network interaction behavior between spam users and legitimate users.In view of this,our article attempts to analyze the abnormal behavior patterns and characteristics of spam users in complex interaction,clarify the role and mechanism of different interaction relationships in identifying spammers,and provide a basis for explaining the importance of interaction behaviors in detecting spam users and how to make full use of these interaction behaviors.Specifically,the main work of this article includes the following two aspects:(1)This article relies on a real social network dataset from Tagged and introduces an e-commerce platform dataset named Yelp Chi as a comparison.The network interaction behavior is analyzed from three aspects: homophily of edges,centrality of nodes,and cohesive groups.Several basic network topology characteristics of networks in two scenarios are quite different.The undirected edges in the co-reviewing network formed by the e-commerce platform have the characteristic of stable homophily,while the directed edges in the social network with multiple interaction relationships do not satisfy this trait.Moreover,spam users are likely to be on the periphery of the network on e-commerce platform,but on the innermost network core on social networking platform.In addition,spam users on e-commerce platform tend to form cohesive groups that are quite scarce on the social networking platform.(2)Through the analysis of the interaction behavior in the social network,it can be found that the topology of the network has a great correlation with the behavior of users,and the interaction behaviors of spam users and legitimate users are also different.The existence of those differences provides us with very useful help in constructing spammer network interaction behavior characteristics and detection models.Based on this,our article constructs a series of interaction sequence features based on interaction time and a series of network topology features based on network structure.Finally,through verification and comparison,the characteristics of network interaction behavior constructed in this article have significantly improved the detection performance of spammers.