Research On Network Structural Regularity Regulation Against Inference Attack

Networks,also called “Graphs”,have been proved to be a powerful representation for different types of datasets with associated relationships,such as communication networks,transportation networks,power networks,social networks,scientific citation networks and so on.With the rapid development of information technology and data collection technology,network data has shown explosive growth,and network data mining has increasingly become the core of numerous important applications.However,different from traditional data,network data has inherent connections and regularity.According to data publishing,malicious attackers can utilize the inherent connections and regularity of network data to inferentially uncover the sensitive relationships of the data generators,which increases the risk of privacy discovery and further hinders network data sharing and utilizing.Therefore,in network data publishing and sharing,how to effectively prevent the privacy threats caused by predition-based inference attack on sensitive relationships has become a common concern in industry and academia.Insipred by the above contradictions,this thesis engages in the study of network structural regularity regulation against inference attacks.The problem aims to achieve privacy-preserving via essential links identification and the structual regulation,which refers to changing the structural patterns of networks to reduce the risk of inference attack based sensitive information disclosure.Specifically,from the level of network inherent regularity,the thesis formulates the structural regularity regulation framework,which integrates network structure modeling,network structural regularity measuring and regulation,and the thesis further analyzes the problem of link centrality measuring.The main work is described as follows:1.In the context of big data analytics,various data mining tasks can effectively promote the development of data analysis applications,but may cause data privacy disclosure problems simultaneously.This thesis designs a framework of network structural regularity regulation to resist inference attacks.Firstly,the microstructure elements identification has been the key foundation of network structural regulation,and this thesis utilizes different link centrality measuring methods to identify representative links of networks.Secondly,based on the effective identification of representative links,the thesis proposes the Network Regularity Regulation mechanism;Finally,taking into account the influence of network regularity regulation behavior,this thesis proposes a low-rank sparse based network structural regularity measuring algorithm,and combines with various network structure prediction methods to jointly establish the evaluation system.2.The existing researchs on link centrality measuring have limitations on the identification of representative links,and thus influences the effect of the network structural regulation.This thesis designs a link importance measuring algorithm.Specifically,based on the idea of random walk,a Random Walk with Indirect Jumps(RWIJ)mechanism is proposed firstly;and according to the theoretical research of network structural regularity,RWIJ based Link Importance Measuring algorithm is proposed.Finally,with the help of the structural regularity regulation framework,the Link Importance Measuring algorithm can be vertified effectively.This thesis utilizes artificial and real network datasets to proves the proposed framework and corresponding algorithms.The experimental results demonstrate the effectiveness of network structural regularity regulation for privacy protection against inference attacks,and prove the RWIJ based link importance measuring algorithm can effectively identify the essential links in the network.