Privacy Enhanced Matrix Factorization for Recommendation with Local Differential Privacy

被引:162
作者
Shin, Hyejin [1 ]
Kim, Sungwook [1 ]
Shin, Junbum [1 ]
Xiao, Xiaokui [2 ]
机构
[1] Samsung Res, Seoul 06765, South Korea
[2] Natl Univ Singapore, Sch Comp, Singapore 119077, Singapore
关键词
Local differential privacy; matrix factorization; recommender system; random projection;
D O I
10.1109/TKDE.2018.2805356
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recommender systems are collecting and analyzing user data to provide better user experience. However, several privacy concerns have been raised when a recommender knows user's set of items or their ratings. A number of solutions have been suggested to improve privacy of legacy recommender systems, but the existing solutions in the literature can protect either items or ratings only. In this paper, we propose a recommender system that protects both user's items and ratings. For this, we develop novel matrix factorization algorithms under local differential privacy (LDP). In a recommender system with LDP, individual users randomize their data themselves to satisfy differential privacy and send the perturbed data to the recommender. Then, the recommender computes aggregates of the perturbed data. This framework ensures that both user's items and ratings remain private from the recommender. However, applying LDP to matrix factorization typically raises utility issues with i) high dimensionality due to a large number of items and ii) iterative estimation algorithms. To tackle these technical challenges, we adopt dimensionality reduction technique and a novel binary mechanism based on sampling. We additionally introduce a factor that stabilizes the perturbed gradients. With MovieLens and LibimSeTi datasets, we evaluate recommendation accuracy of our recommender system and demonstrate that our algorithm performs better than the existing differentially private gradient descent algorithm for matrix factorization under stronger privacy requirements.
引用
收藏
页码:1770 / 1782
页数:13
相关论文
共 29 条
  • [11] Local Privacy and Statistical Minimax Rates
    Duchi, John C.
    Jordan, Michael I.
    Wainwright, Martin J.
    [J]. 2013 IEEE 54TH ANNUAL SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE (FOCS), 2013, : 429 - 438
  • [12] Calibrating noise to sensitivity in private data analysis
    Dwork, Cynthia
    McSherry, Frank
    Nissim, Kobbi
    Smith, Adam
    [J]. THEORY OF CRYPTOGRAPHY, PROCEEDINGS, 2006, 3876 : 265 - 284
  • [13] RAPPOR: Randomized Aggregatable Privacy-Preserving Ordinal Response
    Erlingsson, Ulfar
    Pihur, Vasyl
    Korolova, Aleksandra
    [J]. CCS'14: PROCEEDINGS OF THE 21ST ACM CONFERENCE ON COMPUTER AND COMMUNICATIONS SECURITY, 2014, : 1054 - 1067
  • [14] The MovieLens Datasets: History and Context
    Harper, F. Maxwell
    Konstan, Joseph A.
    [J]. ACM TRANSACTIONS ON INTERACTIVE INTELLIGENT SYSTEMS, 2016, 5 (04)
  • [15] Hua JY, 2015, PROCEEDINGS OF THE TWENTY-FOURTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI), P1763
  • [16] Kairouz P, 2014, ADV NEUR IN, V27
  • [17] Lam SK, 2006, LECT NOTES COMPUT SC, V3995, P14
  • [18] Liu Ziqi, 2015, P 9 ACM C REC SYST, P171, DOI DOI 10.1145/2792838.2800191
  • [19] Personalized Social Recommendations - Accurate or Private?
    Machanavajjhala, Ashwin
    Korolova, Aleksandra
    Das Sarma, Atish
    [J]. PROCEEDINGS OF THE VLDB ENDOWMENT, 2011, 4 (07): : 440 - 450
  • [20] McSherry F, 2009, KDD-09: 15TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, P627