A Method Of Sanitizing A Privacy-sensitive Mobility Knowledge Network Of Trajectory Data Based On A Spark Platform

The mobility rules mined from trajectory data can help to allocate resources and provide personalized services,although these will also pose a threat to privacy of personal locations.The existing methods of eliminating privacy-sensitive mobility patterns mainly adopt the strategy of knowledge hiding.These methods are specified to single types of mining techniques,which include association rule hiding,sequence pattern hiding and so on.However,when a malicious user attacks the structures of mobility knowledge networks with sensitive information,traditional methods cannot effectively deal with inference attacks based on the analysis of mobility knowledge networks.In addition,the current methods of privacy protection based on social networks are mainly targeted at identifying attacks,which are not enough to deal with the mobility inference attacks.To this end,a method of sanitizing sensitive mobility knowledge networks is proposed.In this paper,the trajectory data mined from telecom big data is used as the research object.Firstly,mining mobility patterns and constructing weighted mobility knowledge network are studied.Secondly,a location privacy inference attack model,which includes three attack scenarios based on connectivity analysis,is formally formatted.Thirdly,a method of sanitizing privacy-sensitive mobility knowledge network by identifying and removing influential nodes is designed.Finally,extensive experiments are conducted.The results prove that the interaction between nodes and the closeness of the networks are destroyed,and all the networks maintain the characteristics of “small world” during sanitization.Moreover,the proposed method can completely sanitize the privacy-sensitive mobility knowledge networks from inference attacks.In addition,the proposed method can achieve the optimal values for connectivity degree and security degree by adjusting the parameter of the proportional factors.Furthermore,the performance is shown to be stable for different types of networks and scales for the networks with different sensitivity.Finally,when compared with the method of random deletion,the proportion of nodes to be deleted is smaller in the purification of sensitive rules for complex networks.Therefore,a better hiding effect can be achieved by proposed method.