TY - JOUR
T1 - Protecting multi-party privacy in location-aware social point-of-interest recommendation
AU - Wang, Weiqi
AU - Liu, An
AU - Li, Zhixu
AU - Zhang, Xiangliang
AU - Li, Qing
AU - Zhou, Xiaofang
N1 - KAUST Repository Item: Exported on 2021-02-19
Acknowledgements: Research reported in this publication was partially supported Natural Science Foundation of China (Grant Nos. 61572336, 61572335, 61402313)
PY - 2018/4/4
Y1 - 2018/4/4
N2 - Point-of-interest (POI) recommendation has attracted much interest recently because of its significant business potential. Data used in POI recommendation (e.g., user-location check-in matrix) are much more sparse than that used in traditional item (e.g., book and movie) recommendation, which leads to more serious cold start problem. Social POI recommendation has proved to be an effective solution, but most existing works assume that recommenders have access to all required data. This is very rare in practice because these data are generally owned by different entities who are not willing to share their data with others due to privacy and legal concerns. In this paper, we first propose PLAS, a protocol which enables effective POI recommendation without disclosing the sensitive data of every party getting involved in the recommendation. We formally show PLAS is secure in the semi-honest adversary model. To improve its performance. We then adopt the technique of cloaking area by which expensive distance computation over encrypted data is replaced by cheap operation over plaintext. In addition, we utilize the sparsity of check-ins to selectively publish data, thus reducing encryption cost and avoiding unnecessary computation over ciphertext. Experiments on two real datasets show that our protocol is feasible and can scale to large POI recommendation problems in practice.
AB - Point-of-interest (POI) recommendation has attracted much interest recently because of its significant business potential. Data used in POI recommendation (e.g., user-location check-in matrix) are much more sparse than that used in traditional item (e.g., book and movie) recommendation, which leads to more serious cold start problem. Social POI recommendation has proved to be an effective solution, but most existing works assume that recommenders have access to all required data. This is very rare in practice because these data are generally owned by different entities who are not willing to share their data with others due to privacy and legal concerns. In this paper, we first propose PLAS, a protocol which enables effective POI recommendation without disclosing the sensitive data of every party getting involved in the recommendation. We formally show PLAS is secure in the semi-honest adversary model. To improve its performance. We then adopt the technique of cloaking area by which expensive distance computation over encrypted data is replaced by cheap operation over plaintext. In addition, we utilize the sparsity of check-ins to selectively publish data, thus reducing encryption cost and avoiding unnecessary computation over ciphertext. Experiments on two real datasets show that our protocol is feasible and can scale to large POI recommendation problems in practice.
UR - http://hdl.handle.net/10754/627428
UR - http://link.springer.com/article/10.1007/s11280-018-0550-9
UR - http://www.scopus.com/inward/record.url?scp=85044961977&partnerID=8YFLogxK
U2 - 10.1007/s11280-018-0550-9
DO - 10.1007/s11280-018-0550-9
M3 - Article
AN - SCOPUS:85044961977
SN - 1386-145X
VL - 22
SP - 863
EP - 883
JO - World Wide Web
JF - World Wide Web
IS - 2
ER -