臺灣博碩士論文加值系統

協同過濾是近年來，最被廣泛使用且效果顯著的一類推薦系統模型。然而，其在對於隱私的維護上有很大隱憂，容易受到資料庫漏洞或不被信任的伺服器攻擊。因此，針對這項問題，本篇研究提出一個使用矩陣分解的框架，這個框架乃是利用客戶端上傳梯度的結構，搭配一個兩階段隨機回答的演算法，以此達到對評分的數值、評分的存在以及已訓練之模型的保護。本研究亦使用差異化隱私對此框架具備的隱私程度進行驗證；並透過實驗，成功於數值型回饋任務及單一回饋任務上皆證明其具備一定程度之功用性。

Collaborative filtering (CF) is a popular and widely-used technique for recommendation systems. However, it has privacy concerns of data leakage caused by untrusted servers. To address this problem, we propose a privacy-preserving framework for one of the robustest CF-based method, Matrix Factorization (MF). With the advantage of the characteristic of MF, this framework is based on gradient-transmission client-server architecture to preserve value of feedback and trained model. On basis of this architecture, we further preserve the existence of feedback by a two-stage Randomized Response algorithm. The privacy of this framework is proved to be with the guarantee of differential privacy. We also conduct experiments on numerical feedback task and one-class feedback task. The results demonstrate that our framework can successfully achieve privacy with certain utility.

Acknowledgments i
Abstract iii
List of Figures vii
List of Tables viii
Chapter 1 Introduction 1
Chapter 2 Preliminaries 7
2.1 Preliminaries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7
2.2 Differential Privacy . . . . . . . . . . . . . . . . . . . . . . . . . . . .7
2.3 Matrix Factorization . . . . . . . . . . . . . . . . . . . . . . . . . . .8
Chapter 3 Related Works 11
Chapter 4 Methodology 13
4.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Randomized Response Algorithm . . . . . . . . . . . . . . . . . . . . 14
4.2.1 Permanent Randomized Response (PRR) . . . . . . . . 16
4.2.2 Instantaneous Randomized Response (IRR) . . . . . . . 16
4.3 Computation of Gradients to Unrated Items . . . . . . . . . . . . . . 17
4.4 Control of Differential Privacy . . . . . . . . . . . . . . . . . . . . . . 19
4.5 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Chapter 5 Experiments 24
5.1 Datasets & Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Chapter 6 Conclusions 28
Chapter 7 Future Works 29
Bibliography 30

[1] J. Bennett, S. Lanning, and N. Netflix. The netflix prize. In In KDD Cup and Workshop in conjunction with KDD, 2007.
[2] A. Berlioz, A. Friedman, M. A. Kaafar, R. Boreli, and S. Berkovsky. Applying differential privacy to matrix factorization. In Proceedings of the 9th ACM Conference on Recommender Systems, pages 107–114. ACM, 2015.
[3] J. A. Calandrino, A. Kilzer, A. Narayanan, E. W. Felten, and V. Shmatikov. ” you might also like:” privacy risks of collaborative filtering. In 2011 IEEE Symposium on Security and Privacy, pages 231–246. IEEE, 2011.
[4] J. Canny. Collaborative filtering with privacy. In Security and Privacy, 2002. Proceedings. 2002 IEEE Symposium on, pages 45–57. IEEE, 2002.
[5] C. Dwork, F. McSherry, K. Nissim, and A. Smith. Calibrating noise to sensitivity in private data analysis. In Theory of Cryptography Conference, pages 265–284. Springer, 2006.
[6] C. Dwork, A. Roth, et al. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.
[7]Ú. Erlingsson, V. Pihur, and A. Korolova. Rappor: Randomized aggregatable privacy-preserving ordinal response. In Proceedings of the 2014 ACM SIGSAC conference on computer and communications security, pages 1054–1067. ACM, 2014.
[8] A. Friedman and A. Schuster. Data mining with differential privacy. In Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 493–502. ACM, 2010.
[9] J. Hua, C. Xia, and S. Zhong. Differentially private matrix factorization. In Proceedings of the 24th International Conference on Artificial Intelligence, pages 1763–1770. AAAI Press, 2015.
[10] Y. Koren, R. Bell, C. Volinsky, et al. Matrix factorization techniques for recommender systems. Computer, 42(8):30–37, 2009.
[11] Z. Liu, Y.-X. Wang, and A. Smola. Fast differentially private matrix factorization. In Proceedings of the 9th ACM Conference on Recommender Systems, pages 171–178. ACM, 2015.
[12] A. Machanavajjhala, A. Korolova, and A. D. Sarma. Personalized social recommendations: accurate or private. Proceedings of the VLDB Endowment, 4(7):440–450, 2011.
[13] F. McSherry and I. Mironov. Differentially private recommender systems: building privacy into the net. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 627–636. ACM, 2009.
[14] V. Nikolaenko, S. Ioannidis, U. Weinsberg, M. Joye, N. Taft, and D. Boneh. Privacy-preserving matrix factorization. In Proceedings of the 2013 ACM SIGSAC conference on Computer & communications security, pages 801–812. ACM, 2013.
[15] S. Rendle, C. Freudenthaler, Z. Gantner, and L. Schmidt-Thieme. Bpr: Bayesian personalized ranking from implicit feedback. In Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence, pages 452–461. AUAI Press, 2009.
[16] F. Ricci, L. Rokach, and B. Shapira. Introduction to recommender systems handbook. Springer, 2011.
[17] Y. Shen and H. Jin. Privacy-preserving personalized recommendation: An instance-based approach via differential privacy. In 2014 IEEE International Conference on Data Mining, pages 540–549. IEEE, 2014.
[18] Y. Shen and H. Jin. Epicrec: Towards practical differentially private framework for personalized recommendation. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, pages 180–191. ACM, 2016.
[19] M. Welling and Y. W. Teh. Bayesian learning via stochastic gradient langevin dynamics. In Proceedings of the 28th International Conference on Machine Learning (ICML-11), pages 681–688, 2011.
[20] Y. Xin and T. Jaakkola. Controlling privacy in recommender systems. In Advances in Neural Information Processing Systems, pages 2618–2626, 2014.
[21] G. Yuan, Y. Y. Z. Zhang, and Z. Hao. Convex optimization for linear query processing under approximate differential privacy. arXiv preprint arXiv:1602.04302, 2016.