Research On Privacy-preserving Data Publication For Location-based Recommendation

Big data,the milestone of the information age,is driving a profound change in human society with great power.The development of information technology,such as the Internet and electronic storage,ties up the human and data tightly.In this circumstance,useful data published by data holder can evidently improve knowledge,service and productivity in many social sectors.In the big data time,however,there is another problem viewed as information overloading,namely,an individual has insufficient ability to deal with large amount of information.Recommender system,which provides personalized information for users,give a way to this by sorting to data mining.It is worth mentioning that published data used for recommenddation may breach privacy of individuals.Privacy becomes a barrier in advancement of big data in that some data may explicitly identify an individual.We firstly proposed a scheme on location-based recommendation for points of interest.And then we also propose a privacy-preserving data publishing scheme for location-based recommendation.The main contents of this paper are listed as follows:1.We firstly proposed a scheme on location-based recommendation for points of interest.By analyzing the check-in data,we compute two important parameters,the heat value and the analogue scores,through geological features to obtain the value function.After matching the history preferences of users,we input the distance,which between the user requiring recommendation and alternative POIs,and the value of function into K-dominating query algorithm,choosing the top-K locations sent to users.2.We then propose a scheme on privacy-preserving data publishing for location-based recommendation and display our scheme in detail about generalization of sensitive attribute and trajectory privacy protection with differential privacy.In the generalization part,two methods are exhibited.In the other part,we utilize differential privacy to synthesize a new trajectory which satisfy the requirement of differential privacy according to primary trajectory and sematic class of location.3.Lastly,we analyze the privacy-preserving effect and mechanism performances of our scheme.We exhibit the protection effect through privacy attack models and background knowledge of opponents and compare it with other methods.The analysis of performance includes measurement of sensitive information loss and test of synthetic trajectory in computation and space distortion.Experiments show that our scheme have a better performance whether in privacy protection or in data utility.