Research On Practical Location Privacy Protection For Location-based Services

More and more mobile Applications will access users' locations.This leads to a security issue: malicious Applications may frequently steal or overuse the user's location,which seriously threatens users' privacy.Researches on location security protection have been around for a long time.Unfortunately,by so far,we see few solutions have gotten actual deployment on a large scale.We think the main reason is all existing solutions face a significant contradiction between security and usability.Therefore,this paper will focus on solving the major practical problems of such protection mechanisms.With Geo-indistinguishability,the first strict location privacy model,we may expect an ideal practical location protection tool on mobile systems,which hooks every location request from Apps and perturbs their positioning results mandatorily with a method satisfying this model(e.g.,the planar Laplace mechanism,PLM).This tool requires no cooperation from untrustworthy App providers,which has broad practical prospects.Nevertheless,to put this tool into practice,there are at least two more challenges.First,for a specific location request,how we can automatically identify its fine-grained location precision requirement for producing the downstream functionality.Second,a differential privacy protection model that is suitable for the requirement of relevant accuracy and satisfies the geo-indistinguishability in the scenario of continuous requests from an individual.Location perturbation mechanism PLM is originally designed for protecting a single location.If we independently Apply it to frequent location requests,the privacy costs accumulate quickly,which would soon drain the privacy budget.Only those two points were solved can the user location security protection mechanism implement a fine-grained balance between security and usability.We address both problems in this paper.For the first one,we propose a finegrained precision analysis of location requests.We observe the precision requirement is determined by the downstream service type.We build the first NLP-based service type classification model leveraging service texts statically extracted from APKs.Our experiments show that the average accuracy of the model exceeds 90%.For the second one,we introduce a prediction and test process in PLM that can significantly reduce the real privacy costs most of the time while satisfying the precision requirement and Geo-indistinguishability.Consequently,the speed of privacy budget consumption is slowed down by 70% on average.So that the new mechanism can be better applied to scenarios where Apps requesting locations frequently.