0 introduction

With the rapid development of the Internet and network applications, the network information is growing exponentially, which makes it difficult for users to obtain the useful information efficiently from the massive data.The recommendation system based on collaborative filtering (CF) [1,2] algorithm is an efficient personalized recommendation system. It finds the potential consumption trend of users by mining the relationship between user data, establishes user characteristics, and provides personalized information recommendation for users, which greatly facilitates the information acquisition of users.However, when mining user data, CF based recommendation algorithm has the risk of privacy leakage because the data involved contains user privacy information.How to protect the privacy of users and improve the accuracy of recommendation is the key problem to be solved in the current recommendation system.

Traditional privacy protection technologies, such as k-anonymity, l-identity, t-closeness, etc., rely on specific attack assumptions, and cannot quantify the level of privacy protection.In 2006, dwork et al. Proposed a differential privacy model, which has strict mathematical theory guarantee and provable privacy security.With the difference^[3]

2010 Friedman

Differential privacy is applied to the field of data mining, but it is difficult to allocate the privacy budget by transforming the traditional data mining algorithm into the differential privacy protection algorithm.In recent years, differential privacy protection technology has gradually become the key technology of data privacy protection, and has been widely used.For example, in 2019, Piao et al. Proposed a differential privacy method based on cloud fog computing to protect the data disclosed by the government, which effectively avoided the risk of citizensamp;apos; privacy disclosure in the government data disclosure, but it introduced more noise and reduced the data availability.In 2020, Hou et al. Applied the differential privacy technology to the industrial Internet. By introducing the Top-k frequent

剩余内容已隐藏，支付完成后下载完整资料

Collaborative filtering recommendation algorithm based on hybrid similarity and differential privacy^*

Zhang runlian, Zhang Rui, Wu Xiaonian, Liu wenfen^{a, baaa}

(A. Guangxi Key Laboratory of cryptography and information security, Guilin University of Electronic Science and technology; B. Guangxi Key Laboratory of cloud computing and complex systems, Guilin, Guangxi

541004)

Absrtact: in the existing collaborative filtering recommendation algorithms, some factors such as one-sided rating, strong subjectivity, sparsity of rating matrix affect the accuracy of recommendation, and there are privacy leakage problems in recommendation.To solve these problems, a collaborative filtering recommendation algorithm based on hybrid similarity and differential privacy is proposed.In order to improve the accuracy of recommendation, the hybrid similarity is used as the centroid update and classification condition, and the improved K-means is usedThe algorithm clusters the users with high similarity with the target users; uses enumeration method to divide the target user set into subsets, and constructs utility function based on mixed similarity, uses differential privacy index mechanism to select neighbor sets in each subset to protect user privacy; finally, selects the item with the highest score in the neighbor set for recommendation.The experimental results show that the algorithm can effectively protect the privacy of users and improve the accuracy of recommendation.

Keywords: recommender system; privacy protection; collaborative filtering algorithm; differential privacy; hybrid similarity

Chinese Library Classification No.: TP doi: 10.19734/j.issn.1001-3695.2020.12.0542

Collaborative filtering recommendation algorithm based on mixed similarity and differential privacy

Zhang Runliana, b, Zhang Ruia, Wu Xiaoniana, Liu Wenfena

(a. Guangxi Key Laboratory of Cryptography amp; Information Security, b. Guangxi Colleges Key Laboratory of Cloud

Computingamp; Complex Systems, Guilin University of Electronic Technology, Guilin Guangxi 541004, China)

Abstract:The existing collaborative filtering recommendation algorithms have some problems. Some factors, such as onesided rating, strong subjectivity, sparsity rating matrix, and so on, affect the accuracy of recommendation. Moreover, there is privacy leakage in the recommendation. To solve the above problems, this paper proposes a collaborative filtering recommendation algorithm based on the mixed similarity and differential privacy. Firstly, in order to improve the recommendation accuracy, this paper construct the mixed similarity based on the weighted calculation for some similarity methods. Then, the mixed similarity is used as the condition for the centroid updating and classification of the K-means algorithm, and users with high similarity to target user are clustered by the improved K-means algorithm. Furthermore, the algorithm divides the target users into subsets by the enumeration method, and constructs the utility function based on the mixed similarity. Specifically, for purpose of protecting usersamp;apos; privacy, based on the differential privacy indexing mechanism of utility function, the algorithm selects the neighbor sets from subsets. Finally, the algorithm selects and recommends the items with the highest score from the neighbor sets. The experimental results show that the proposed algorithm can not only protect usersamp;apos; privacy, but also improve the accuracy of recommendation effectively.

Key words:recommendation system; privacy protect; collaborative filtering; differential privacy; mixed similarity