Research On Privacy Protection Methods For Data Publishing And Sharing

With the emerging rise and development of big data,cloud computing and other technologies,the issue of data privacy protection has become a research hot spot in the field of network and information security.In recent years,privacy protection technologies for data release and sharing have emerged and been widely used in the public domain of society.However,existing privacy-preserving technologies still have obvious defects:first,they cannot effectively balance the relationship between data privacy and data availability,second,they cannot predict the dependency relationship between data attributes among data owners in multi-source data fusion scenarios,as well as cannot effectively improve the usability of fused data,and third,they cannot guarantee that the privacy-preserved data can still be effectively data mined.In this paper,the research is closely focused on these three aspects of the problem,based on Bayesian networks,collaborative filtering,and reinforcement learning,we construct a privacy-preserving architecture for data publishing and sharing,and propose a targeted privacy-preserving data publishing method based on Bayesian networks,a combined privacy-preserving policy optimization method for multi-source data fusion,and a privacy-preserving data publishing method based on reinforcement learning.The specific work is as follows.First,to address the inability to effectively balance between data privacy and data availability in complex data publishing scenarios,this paper designs a privacy-preserving data publishing method based on Bayesian networks,constructs a re-anonymization privacy-preserving architecture,and adopts a Bayesian network structure learning method and a migration method for privacy-preserving operations to achieve internal and external anonymity of data.In addition,to address the lack of targeting of the above anonymization methods,the attribute filtering technique based on the discriminative matrix is proposed,the discriminative matrix and attribute hypergraph are constructed,the heuristic method of attribute hypergraph disambiguation is adopted to maximize the contribution of attribute value differences to sensitive information mining,and the targeting of privacy protection is achieved,and then this paper also proposes the（d,（?））-perturbation mechanism,which prevents excessive modification of a core information attribute.The simulation experimental results show that the Bayesian network-based targeted privacypreserving data publishing method proposed in this paper can improve the usability of data while satisfying the data privacy.Second,to address the problem that it is difficult to predict the dependency relationship between data attributes in the scenario of multi-source data fusion publishing,this paper designs a privacy-preserving method for multi-source data fusion,constructs a system model for multi-source data fusion,proposes a Bayesian network fusion algorithm based on collaborative filtering,and adopts a matrix decomposition method to achieve the prediction of the dependency relationship between attributes of different data sets,which effectively prevents the fusion data privacy leakage.In addition,to address the problem of low availability of fused data,this paper designs heuristic algebraic rules,proposes a multi-strategy combination optimization algorithm for multi-source data fusion,and adopts the technique of hypergraph disambiguation to optimize the process of multi-privacy protection strategy fusion,which achieves the purpose of improving the availability of fused data.Simulation experiments on three different types of datasets show that the mean square error of the prediction of the Bayesian network fusion method based on collaborative filtering proposed in this paper stays within 0.1 and has strong robustness on medium and large datasets,and in addition,the data availability of the multi-policy combination optimization algorithm for multi-source data fusion with different parameter settings is higher than that of the other three benchmark algorithms,indicating that this algorithm can The availability of fused data is greatly improved.Third,to address the problem that the serial fragmentation of privacy protection operation and data mining operation leads to the reduction of data utility,this paper proposes a reinforcement learning-based privacy protection data publishing method,constructs a role-based interaction model,evolves a privacy protection intelligent body and a data mining intelligent body,designs a process for the privacy protection intelligent body to modify Bayesian networks and an integrated clustering process for the data mining intelligent body,and adopts a deep reinforcement learning approach to establish a game between the two intelligences,optimize the interaction process between them,and achieve a synergistic balance between privacy-preserving and data mining.The experimental results of comparison with two benchmark algorithms show that the algorithm proposed in this paper can maximize the gain in a limited time and achieve privacy protection while ensuring the effectiveness of data mining.