| dc.description.abstract | Recently, location-based social networks (LBSNs), such as Foursquare, Facebook, and Plenty of Fish, have attracted millions of users by helping them to build their social contacts and share useful information. LBSNs provide fine-grained and personalized services to their users, such as the 'location check-in' feature in check-in applications to obtain rewards, the 'people nearby' feature in dating applications for meeting interested people nearby, and the 'friends alert' feature in proximity notification applications for receiving a notice when a friend is close by. Users can use these features to do activities such as a) update their friends on their whereabouts; b) discover the best places to eat, drink, shop, or visit in a certain area; c) find new friends with similar interests or a date with a well matched person; and d) easily organize social activities. However, location sharing is a double-edged sword, which can on the one hand make life convenient but also might reveal private location information to curious application servers or malicious users.
In state-of-the-art location privacy mechanisms, some only provide location privacy protection against malicious users but not curious application servers. While some of the existing solutions obfuscate the actual location of their users, such as cell-grid based approaches, they suffer from false positives/negatives in proximity estimation. Some other solutions implement stronger security, i.e., location tags, to protect users' location privacy. However, these techniques usually rely on additional sniffing devices that are costly and cause an excessive drain on the battery of the users' handsets. The existing mechanisms are only designed to address privacy issues for specific types of applications in LBSNs and all suffer from different types of privacy attacks.
Our research contributions address four aspects relating to location privacy:
1. We address issues of privacy in LBSNs by protecting user's privacy against both application servers and malicious users; our approach employs encrypted cell-tower identifier sets instead of location coordinates for location proximity testing, which protects users' actual locations from being revealed to either malicious users or application servers.
2. We create a privacy-aware mechanism that satisfies all categories of LBSN applications. For instance, 'k-anonymity' is designed to hide users' identities and actual locations when they send queries to fetch information from servers. This approach could be used in check-in applications but not dating and proximity applications that need to show users' identities. Our approach doesn't share actual location coordinates with any party but allows users or application servers to achieve application functions related to location.
3. We employ cell-tower identifiers as location tags, which can be directly accessed by mobile phones without a sniffing tool. This solution is able to resist more privacy attacks with an encrypted dataset and is more mobile friendly. Cell towers are distributed according to the density of mobile users, and their coverage ranges are also adjusted according to the density of mobile users. This means that a group of cell towers that covers a mobile user would dynamically shape a particular obfuscated region for each different location to which the mobile users move. Hence, we adopt this feature to provide a self-organizing location obfuscation solution to ensure users' location information is protected in LBSN applications.
4. We introduce the k-combination approach which is a more accurate proximity testing mechanism than k-shingling approach. Our approach takes all similar elements between two data sets into account when comparing the similarity of those data sets. Our experiments show that the results using our k-combination approach have better accuracy when compared with data using the k-shingling approach. | en |
