Tensor Ring Decomposition Based Collaborative Filtering Recommendation with Differential Privacy

Recommendation systems offer a solution to problems such as information overload and lack of relevant information, thereby enhancing sales and improving users satisfaction. Collaborative filtering (CF) uses the preferences of a known user group to predict the preferences of unknown users. At present, many related models have been proposed, with matrix and CP factorizations being particularly popular. But in practical recommendation applications, matrix decomposition is difficult to deal with high-dimensional data, and CP decomposition is limited by its few parameters. To optimize the efficacy of recommendations while still retaining a low level of complexity, we propose employing a tensor ring (TR) decomposition framework for CF recommendation. In this paper, TR decomposition the original tensor to obtain core tensors and a rank vector. To minimize the rating error, a loss function is constructed, and stochastic gradient descent is used to repeatedly update each parameter. In addition, to protect user sensitive information, we employ Laplace random noise to disturbe ratings. By controlling the value of the privacy budget, a balance is achieved between the preservation of data privacy and the utility of data. Extensive experiments show that our approach not only improves recommendations but also protects users privacy by using this perturbation technique.