USLD: A New Approach for Preserving Location Privacy in LBS

Abstract

The development of LBS bring great convenience to our lives, but also presents new challenges to privacy protection. Many of the existing methods are inadequate because in their schemes they assume that all users can be trusted which is not practical. So, the existing methods can not resist the query sampling attack and self-betrayal attacks. In addition to this, they also did not take the location semantic into account, so they are vulnerable to location homogeneity attacks. In order to solve the problems, we introduce the concept of USLD (user similar location diversity), we consider the scenario that part of the users are not trusted, and the users which we choose as candidates may in the locations which have same semantic. We consider some of users to be untrustworthy, propose the idea that users who have similar privacy settings with the real user are more plausible than others. We select users who are similar with the real user using Adjusted Cosine Similarity, and the Earth Mover Distance is used to calculate location semantics. Our method can well resist query sampling attacks, self-betrayal attacks, location homogeneity attacks. Experiments show that our method is very practical.