Privacy Inference Technology And Application Based On Graph Data

The social activities of users on the Internet can be represented as graph data,the so-called graph data is to abstract the generated data onto graphs,where nodes represent individuals and edges represent the relationships between nodes.Graph data has been widely used in data mining,intelligent recommendation,fraud detection and other network services.In view of this,graph data plays an important supporting role in improving network information services.However,due to the privacy inference attack of the adversary,the graph data have the problem of privacy leakage.The deanonymization attack on graph data is a typical privacy inference attack method.The existing work has limitations in protecting the identity and attributes of graph nodes.The paper focuses on the privacy inference problem of anonymous graph data.From the perspective of privacy attack,and through the study of graph data privacy quantification,the contradiction between graph privacy and utility is solved.The specific research worksare as follows:(1)A privacy inference method for graph data attribute privacy based on probability mapping is proposed.In the view of graph privacy attacks and anonymous privacy protection methods of social network,datamining association analysis method is used to quantify attribute correlation withanalysis the technical principles of the above methods.Furthermore,proposed an attribute privacy inference method based on network topology structure through using the theory and method of probability graph model.(2)According to the proposed attribute privacy inference method,oriented the graph attribute privacy inference algorithm was designed.The FP-Growth algorithm is used to analyze the correlation between node attributes to the mining graph data structure,establish the probabilistic mapping among graph attributes,and realize the de-anonymization of anonymous graph data.Based on this,an attribute privacy inference algorithm is designed through the calculation of graph structure feature similarity and attribute value to realize the probabilistic calculation of user privacy inference.The experimental results show that the designed algorithm has obvious advantages over the existing algorithms in terms of information acquisition of privacy inference,which effectively guarantees the efficiency of inference.(3)Designed and implemented a prototype privacy inference systemfor graph data attribute.A prototype system for privacy inference of graph data attributes has implemented depend on the flowing,such as proposed scheme and the designed algorithm,the software design and development stages,using object-oriented programming language and Prefuse framework,interface-oriented programming and so on.The system includes functional modules such as data preprocessing,attributeprivacy inference,privacy leakage measurement and data visualization.According to the software testing method,the designed and implemented system have achieved the expected functional requirements.