融合标签相似度的差分隐私矩阵分解推荐算法

推荐系统需要利用到大量的用户行为数据，这些数据极有可能暴露用户的喜好，给人们关心的隐私问题带来巨大的挑战。为保证推荐精度与用户隐私，提出一种结合差分隐私与标签信息的矩阵分解推荐模型。该模型首先将标签信息加入到项目相似度的计算过程；随后融入到矩阵分解推荐模型中提高推荐精度；最后运用随机梯度下降法求解模型最优值。为解决用户隐私问题，将拉普拉斯噪声划分成两部分，分别加入项目相似度与梯度求解过程中，使得整个推荐过程满足ε-差分隐私，并在一个真实的数据集上分析验证算法的有效性。实验表明，提出的方法能在保证用户隐私的情况下，仍具有较高的推荐精度。

Differential privacy matrix factorization recommendation algorithm fusing tag similarity

The recommendation system needs to utilize a large amount of user data, which may expose the user''s preferences and pose a huge challenge to the privacy concerns. To ensure the accuracy of recommendation and user privacy, this paper proposed a matrix factorization recommendation model combining differential privacy and tag information. Firstly, the model added the tag information to the process of calculating item similarity, then integrated it into the recommendation model to improve the recommendation accuracy. Finally, this paper solved the model optimal value by the stochastic gradient descent method. For protecting users from privacy threats, the proposed approach divided Laplace noise into two parts, which were added to the process of item similarity and gradient solution respectively, so that the whole recommendation process satisfied the differential privacy, and analyzed the validity of the algorithm on a real data set. Experimental results show that the proposed method has high recommendation accuracy while protecting users'' privacy.