Research On Mechanisms Of Trusted Identity Management And Monitoring In Online Social Network

With the development of information technologies,online social networks(OSNs)have become the main platforms for communication and information sharing.Various kinds of OSNs are penetrating into people’s work and life,combing the online life and offline life closer.This combination has brought about an evolution to network identities,and promoted the development of trusted identities.The trusted identity depends on the consistency between online social network identity and real life identity.If they are presented exactly the same,the network identity is trusted.Otherwise,it’s partially trusted or untrusted.Constructing trusted network identity is helpful not only to enhance user trust and clear network environment,but also to restraint misbehaviors and fight against network crimes.However,due to the original design of Internet,which is open and anonymous,the deceptive and arbitrary use of fake identity has become pervasive.What’s more,spammers and other untrusted network groups use online social netwroks for advertisement abusing,malware distribution,network fraud and rumor spread.They cause serious pollution to network environment and threats to network security.Even worser,they may mislead the public opinion and cause social upheaval and social panic.Even though the OSNs are requiring trusted identities,there’re no correlative supporting mechanisms.In this situation,it’s critical to provide mechanisms for management of trusted identities and detection of untrusted ones.In an attempt to deal with the problems brought about by untrusted identities,we conduct deep researches in identity management and detection.Identity management is used to provide mechanisms for trust assurance.A novel identity model is designed and studied for the conditions of constructing trusted identity.A verification mechanism,which hides the real-name from the server,is proposed for identity verification.Identity detection is used to discover untrusted identities that cause security threats.Considering the bad effects of spammers,we proposed influence-based,trust relation-based and interaction-based methods for spammer detection.Considering the present situation that various kinds of OSNs are isolated with each other,making the OSNs unconnected and the network identities irrelevant,a novel joint model is proposed to characterize the identities in corss-domains.We treat the real world and each OSN as a specific domain.Every user presents his inner-domain identity in each domain.Formalizations of the components of inner-domain identity and their relations are given to analyze the reasons of the emergences of trusted and untrusted identities.Two methods,personal property-based and social property-based,are proposed to solve the problem of identity verification.A prototype system is designed on the basis of physical domain and online social network domain.The experimental result shows that our model can effectively provide supports to the construction of trusted identity.In order to deal with the privacy leaking problem of real name policy,a network identity verification mechanism is built on the basis of trust relationship.In many OSNs,the relations can always map to the relations in real world,where the trust relations are always built on real identities.If a user wants to verify his identity,first he chooses a number of friends as his verifiers.Then verification requests are sent to the server with his network identity,and to the verifiers with his real life identity.The verifiers check the real name in the request and submit the results to the server with their own network identities.Finally,the server only need to store the verify relations rather than the real names,which avoids privacy leaking.Some users are request to verify with their real names for the purpose of user tracing and avoiding mutual verification of small groups.These users are treated as roots to verify others,and only the ones that have been verified are qualified to be verifiers.Our method can achieve similar effects with real name policy,besides,it avoids the problem of privacy leaking.An influence-based detection method is proposed for spammer detection.First the characters of influence are analyzed.The influence is measured as the combination of a long-term influence and a short-term influence.We conduct deep research in the burst property of influence,finding that the burst change of normal user’s influence mostly results from the change of interaction group,which will affect the long-term influence.While the burst change of spammer’s influence is seldom accompanied by the change of interaction group,which only affects the short-term influence.With a continual measurement and analysis of long-term and short-term influence,we are able to detect spammers effectively.Our detection method is verified by the experiment with data from micro-blogging of Sina.A relation-based detection method is proposed to detect Sybil groups in online social networks.Sybil group is a type of spammers.Users inside the group are tightly connected,while loosely connected with users outside the group.We define a network with the trust relations,and use path as factor of trust measurement.A path announcement procedure and a path aggregation procedure are conducted to evaluate user trust.The users in Sybil groups are distinguished with low trusts.Traditional detection methods suffer from the bridge problem,which implies that some members in Sybil group may share paths with others to avoid detection.We define K-similar paths and propose a path verification mechanism to reduce the effects of sharing and fake paths,which can conquer the bridge problem effectively.An interaction-based method is proposed for spammer detection.Repost and comment are main reasons that cause information propagation.Spammers are hired to repost and comment for financial benefits,while normal users repost and comment for communication and information sharing.This difference causes different behaviors.For example,the actions of spammers are more concentrated and last shorter,which provides clues for detection.We define six properties to construct a decision tree for spammer detection.Different from the traditional detection methods such as content-based,which is only suitable to a single type of spammers,our method is effective for various kinds of spammers.The reason lies that although there’re many kinds of spammers,they behave the same way when the spam campaign is launched.A prototype system is built based on the technologies above,which can provide monitoring and measurement to identities in online social networks.Our system analyzes users’ identity properties from the data from all kinds of online social networks.User trusts are analyzed and measured to provide support for constructing more trusted network identities.